mirror of
https://github.com/denizsafak/abogen.git
synced 2026-07-18 05:40:26 +02:00
chore: normalize line endings to LF, add .gitattributes
- Add .gitattributes with text=auto eol=lf for all text files - Renormalize all files in index to LF line endings - Fixes massive whitespace-only diffs between main and feature branch
This commit is contained in:
Artem Akymenko
+177
-177
@@ -1,178 +1,178 @@
|
||||
"""Kokoro TTS Plugin for the TTS Plugin Architecture.
|
||||
|
||||
This plugin provides a Kokoro-based TTS engine that implements the
|
||||
Plugin API contract. It wraps the existing Kokoro backend in the
|
||||
new Engine/EngineSession architecture.
|
||||
|
||||
Exports:
|
||||
- PLUGIN_MANIFEST: PluginManifest
|
||||
- MODEL_REQUIREMENTS: list[ModelManifest]
|
||||
- create_engine: Factory function
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from abogen.tts_plugin.engine import Engine
|
||||
from abogen.tts_plugin.host_context import HostContext
|
||||
from abogen.tts_plugin.manifest import (
|
||||
AudioFormatManifest,
|
||||
EngineManifest,
|
||||
ModelManifest,
|
||||
ParameterManifest,
|
||||
PluginManifest,
|
||||
RequirementManifest,
|
||||
VoiceManifest,
|
||||
VoiceSourceManifest,
|
||||
"""Kokoro TTS Plugin for the TTS Plugin Architecture.
|
||||
|
||||
This plugin provides a Kokoro-based TTS engine that implements the
|
||||
Plugin API contract. It wraps the existing Kokoro backend in the
|
||||
new Engine/EngineSession architecture.
|
||||
|
||||
Exports:
|
||||
- PLUGIN_MANIFEST: PluginManifest
|
||||
- MODEL_REQUIREMENTS: list[ModelManifest]
|
||||
- create_engine: Factory function
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from abogen.tts_plugin.engine import Engine
|
||||
from abogen.tts_plugin.host_context import HostContext
|
||||
from abogen.tts_plugin.manifest import (
|
||||
AudioFormatManifest,
|
||||
EngineManifest,
|
||||
ModelManifest,
|
||||
ParameterManifest,
|
||||
PluginManifest,
|
||||
RequirementManifest,
|
||||
VoiceManifest,
|
||||
VoiceSourceManifest,
|
||||
)
|
||||
from abogen.tts_plugin.types import EngineConfig
|
||||
|
||||
from .engine import KokoroEngine
|
||||
|
||||
|
||||
def _load_kpipeline() -> Any:
|
||||
"""Lazy-load Kokoro dependencies."""
|
||||
from kokoro import KPipeline # type: ignore[import-not-found]
|
||||
return KPipeline
|
||||
|
||||
|
||||
PLUGIN_MANIFEST = PluginManifest(
|
||||
id="kokoro",
|
||||
name="Kokoro",
|
||||
version="0.9.4",
|
||||
api_version="1.0",
|
||||
description="Kokoro TTS engine - high quality multilingual text-to-speech",
|
||||
author="Kokoro Team",
|
||||
capabilities=("voice_list",),
|
||||
requires=RequirementManifest(
|
||||
internet=False,
|
||||
),
|
||||
engine=EngineManifest(
|
||||
voiceSources=(
|
||||
VoiceSourceManifest(
|
||||
id="builtin",
|
||||
name="Built-in Voices",
|
||||
type="list",
|
||||
config={"voices": "See listVoices()"},
|
||||
),
|
||||
),
|
||||
parameters=(
|
||||
ParameterManifest(
|
||||
id="speed",
|
||||
name="Speed",
|
||||
description="Speech speed multiplier",
|
||||
type="float",
|
||||
default=1.0,
|
||||
min=0.5,
|
||||
max=2.0,
|
||||
step=0.1,
|
||||
),
|
||||
),
|
||||
audioFormats=(
|
||||
AudioFormatManifest(mime="audio/wav", extension="wav"),
|
||||
),
|
||||
),
|
||||
voices=(
|
||||
VoiceManifest(id="af_alloy", name="Alloy", tags=("en", "female")),
|
||||
VoiceManifest(id="af_aoede", name="Aoede", tags=("en", "female")),
|
||||
VoiceManifest(id="af_bella", name="Bella", tags=("en", "female")),
|
||||
VoiceManifest(id="af_heart", name="Heart", tags=("en", "female")),
|
||||
VoiceManifest(id="af_jessica", name="Jessica", tags=("en", "female")),
|
||||
VoiceManifest(id="af_kore", name="Kore", tags=("en", "female")),
|
||||
VoiceManifest(id="af_nicole", name="Nicole", tags=("en", "female")),
|
||||
VoiceManifest(id="af_nova", name="Nova", tags=("en", "female")),
|
||||
VoiceManifest(id="af_river", name="River", tags=("en", "female")),
|
||||
VoiceManifest(id="af_sarah", name="Sarah", tags=("en", "female")),
|
||||
VoiceManifest(id="af_sky", name="Sky", tags=("en", "female")),
|
||||
VoiceManifest(id="am_adam", name="Adam", tags=("en", "male")),
|
||||
VoiceManifest(id="am_echo", name="Echo", tags=("en", "male")),
|
||||
VoiceManifest(id="am_eric", name="Eric", tags=("en", "male")),
|
||||
VoiceManifest(id="am_fenrir", name="Fenrir", tags=("en", "male")),
|
||||
VoiceManifest(id="am_liam", name="Liam", tags=("en", "male")),
|
||||
VoiceManifest(id="am_michael", name="Michael", tags=("en", "male")),
|
||||
VoiceManifest(id="am_onyx", name="Onyx", tags=("en", "male")),
|
||||
VoiceManifest(id="am_puck", name="Puck", tags=("en", "male")),
|
||||
VoiceManifest(id="am_santa", name="Santa", tags=("en", "male")),
|
||||
VoiceManifest(id="bf_alice", name="Alice", tags=("en", "female")),
|
||||
VoiceManifest(id="bf_emma", name="Emma", tags=("en", "female")),
|
||||
VoiceManifest(id="bf_isabella", name="Isabella", tags=("en", "female")),
|
||||
VoiceManifest(id="bf_lily", name="Lily", tags=("en", "female")),
|
||||
VoiceManifest(id="bm_daniel", name="Daniel", tags=("en", "male")),
|
||||
VoiceManifest(id="bm_fable", name="Fable", tags=("en", "male")),
|
||||
VoiceManifest(id="bm_george", name="George", tags=("en", "male")),
|
||||
VoiceManifest(id="bm_lewis", name="Lewis", tags=("en", "male")),
|
||||
VoiceManifest(id="ef_dora", name="Dora", tags=("es", "female")),
|
||||
VoiceManifest(id="em_alex", name="Alex", tags=("es", "male")),
|
||||
VoiceManifest(id="em_santa", name="Santa", tags=("es", "male")),
|
||||
VoiceManifest(id="ff_siwis", name="Siwis", tags=("fr", "female")),
|
||||
VoiceManifest(id="hf_alpha", name="Alpha", tags=("hi", "female")),
|
||||
VoiceManifest(id="hf_beta", name="Beta", tags=("hi", "female")),
|
||||
VoiceManifest(id="hm_omega", name="Omega", tags=("hi", "male")),
|
||||
VoiceManifest(id="hm_psi", name="Psi", tags=("hi", "male")),
|
||||
VoiceManifest(id="if_sara", name="Sara", tags=("it", "female")),
|
||||
VoiceManifest(id="im_nicola", name="Nicola", tags=("it", "male")),
|
||||
VoiceManifest(id="jf_alpha", name="Alpha", tags=("ja", "female")),
|
||||
VoiceManifest(id="jf_gongitsune", name="Gongitsune", tags=("ja", "female")),
|
||||
VoiceManifest(id="jf_nezumi", name="Nezumi", tags=("ja", "female")),
|
||||
VoiceManifest(id="jf_tebukuro", name="Tebukuro", tags=("ja", "female")),
|
||||
VoiceManifest(id="jm_kumo", name="Kumo", tags=("ja", "male")),
|
||||
VoiceManifest(id="pf_dora", name="Dora", tags=("pt", "female")),
|
||||
VoiceManifest(id="pm_alex", name="Alex", tags=("pt", "male")),
|
||||
VoiceManifest(id="pm_santa", name="Santa", tags=("pt", "male")),
|
||||
VoiceManifest(id="zf_xiaobei", name="Xiaobei", tags=("zh", "female")),
|
||||
VoiceManifest(id="zf_xiaoni", name="Xiaoni", tags=("zh", "female")),
|
||||
VoiceManifest(id="zf_xiaoxiao", name="Xiaoxiao", tags=("zh", "female")),
|
||||
VoiceManifest(id="zf_xiaoyi", name="Xiaoyi", tags=("zh", "female")),
|
||||
VoiceManifest(id="zm_yunjian", name="Yunjian", tags=("zh", "male")),
|
||||
VoiceManifest(id="zm_yunxi", name="Yunxi", tags=("zh", "male")),
|
||||
VoiceManifest(id="zm_yunxia", name="Yunxia", tags=("zh", "female")),
|
||||
VoiceManifest(id="zm_yunyang", name="Yunyang", tags=("zh", "male")),
|
||||
),
|
||||
)
|
||||
|
||||
MODEL_REQUIREMENTS: list[ModelManifest] = []
|
||||
|
||||
|
||||
def create_engine(
|
||||
context: HostContext,
|
||||
model_path: Path | None,
|
||||
config: EngineConfig,
|
||||
) -> Engine:
|
||||
"""Create a Kokoro engine instance.
|
||||
|
||||
This function is the plugin entry point. It must be atomic:
|
||||
succeed fully or raise EngineError and clean up.
|
||||
|
||||
Args:
|
||||
context: Host services (config dir, logger, http client).
|
||||
model_path: Resolved model path, or None for default.
|
||||
config: Engine initialization settings (device, etc.).
|
||||
|
||||
Returns:
|
||||
A fully initialized KokoroEngine instance.
|
||||
|
||||
Raises:
|
||||
EngineError: On failure. Cleans up partially created resources.
|
||||
"""
|
||||
try:
|
||||
KPipeline = _load_kpipeline()
|
||||
|
||||
# Determine repo_id from model_path or use default
|
||||
repo_id = "hexgrad/Kokoro-82M"
|
||||
if model_path is not None:
|
||||
# If a specific model path is provided, use it as repo_id
|
||||
repo_id = str(model_path)
|
||||
|
||||
pipeline = KPipeline(
|
||||
lang_code=config.lang_code,
|
||||
repo_id=repo_id,
|
||||
device=config.device,
|
||||
)
|
||||
|
||||
engine = KokoroEngine(pipeline)
|
||||
return engine
|
||||
except Exception as e:
|
||||
from abogen.tts_plugin.errors import EngineError as EngineErrorClass
|
||||
raise EngineErrorClass(f"Failed to create Kokoro engine: {e}") from e
|
||||
from abogen.tts_plugin.types import EngineConfig
|
||||
|
||||
from .engine import KokoroEngine
|
||||
|
||||
|
||||
def _load_kpipeline() -> Any:
|
||||
"""Lazy-load Kokoro dependencies."""
|
||||
from kokoro import KPipeline # type: ignore[import-not-found]
|
||||
return KPipeline
|
||||
|
||||
|
||||
PLUGIN_MANIFEST = PluginManifest(
|
||||
id="kokoro",
|
||||
name="Kokoro",
|
||||
version="0.9.4",
|
||||
api_version="1.0",
|
||||
description="Kokoro TTS engine - high quality multilingual text-to-speech",
|
||||
author="Kokoro Team",
|
||||
capabilities=("voice_list",),
|
||||
requires=RequirementManifest(
|
||||
internet=False,
|
||||
),
|
||||
engine=EngineManifest(
|
||||
voiceSources=(
|
||||
VoiceSourceManifest(
|
||||
id="builtin",
|
||||
name="Built-in Voices",
|
||||
type="list",
|
||||
config={"voices": "See listVoices()"},
|
||||
),
|
||||
),
|
||||
parameters=(
|
||||
ParameterManifest(
|
||||
id="speed",
|
||||
name="Speed",
|
||||
description="Speech speed multiplier",
|
||||
type="float",
|
||||
default=1.0,
|
||||
min=0.5,
|
||||
max=2.0,
|
||||
step=0.1,
|
||||
),
|
||||
),
|
||||
audioFormats=(
|
||||
AudioFormatManifest(mime="audio/wav", extension="wav"),
|
||||
),
|
||||
),
|
||||
voices=(
|
||||
VoiceManifest(id="af_alloy", name="Alloy", tags=("en", "female")),
|
||||
VoiceManifest(id="af_aoede", name="Aoede", tags=("en", "female")),
|
||||
VoiceManifest(id="af_bella", name="Bella", tags=("en", "female")),
|
||||
VoiceManifest(id="af_heart", name="Heart", tags=("en", "female")),
|
||||
VoiceManifest(id="af_jessica", name="Jessica", tags=("en", "female")),
|
||||
VoiceManifest(id="af_kore", name="Kore", tags=("en", "female")),
|
||||
VoiceManifest(id="af_nicole", name="Nicole", tags=("en", "female")),
|
||||
VoiceManifest(id="af_nova", name="Nova", tags=("en", "female")),
|
||||
VoiceManifest(id="af_river", name="River", tags=("en", "female")),
|
||||
VoiceManifest(id="af_sarah", name="Sarah", tags=("en", "female")),
|
||||
VoiceManifest(id="af_sky", name="Sky", tags=("en", "female")),
|
||||
VoiceManifest(id="am_adam", name="Adam", tags=("en", "male")),
|
||||
VoiceManifest(id="am_echo", name="Echo", tags=("en", "male")),
|
||||
VoiceManifest(id="am_eric", name="Eric", tags=("en", "male")),
|
||||
VoiceManifest(id="am_fenrir", name="Fenrir", tags=("en", "male")),
|
||||
VoiceManifest(id="am_liam", name="Liam", tags=("en", "male")),
|
||||
VoiceManifest(id="am_michael", name="Michael", tags=("en", "male")),
|
||||
VoiceManifest(id="am_onyx", name="Onyx", tags=("en", "male")),
|
||||
VoiceManifest(id="am_puck", name="Puck", tags=("en", "male")),
|
||||
VoiceManifest(id="am_santa", name="Santa", tags=("en", "male")),
|
||||
VoiceManifest(id="bf_alice", name="Alice", tags=("en", "female")),
|
||||
VoiceManifest(id="bf_emma", name="Emma", tags=("en", "female")),
|
||||
VoiceManifest(id="bf_isabella", name="Isabella", tags=("en", "female")),
|
||||
VoiceManifest(id="bf_lily", name="Lily", tags=("en", "female")),
|
||||
VoiceManifest(id="bm_daniel", name="Daniel", tags=("en", "male")),
|
||||
VoiceManifest(id="bm_fable", name="Fable", tags=("en", "male")),
|
||||
VoiceManifest(id="bm_george", name="George", tags=("en", "male")),
|
||||
VoiceManifest(id="bm_lewis", name="Lewis", tags=("en", "male")),
|
||||
VoiceManifest(id="ef_dora", name="Dora", tags=("es", "female")),
|
||||
VoiceManifest(id="em_alex", name="Alex", tags=("es", "male")),
|
||||
VoiceManifest(id="em_santa", name="Santa", tags=("es", "male")),
|
||||
VoiceManifest(id="ff_siwis", name="Siwis", tags=("fr", "female")),
|
||||
VoiceManifest(id="hf_alpha", name="Alpha", tags=("hi", "female")),
|
||||
VoiceManifest(id="hf_beta", name="Beta", tags=("hi", "female")),
|
||||
VoiceManifest(id="hm_omega", name="Omega", tags=("hi", "male")),
|
||||
VoiceManifest(id="hm_psi", name="Psi", tags=("hi", "male")),
|
||||
VoiceManifest(id="if_sara", name="Sara", tags=("it", "female")),
|
||||
VoiceManifest(id="im_nicola", name="Nicola", tags=("it", "male")),
|
||||
VoiceManifest(id="jf_alpha", name="Alpha", tags=("ja", "female")),
|
||||
VoiceManifest(id="jf_gongitsune", name="Gongitsune", tags=("ja", "female")),
|
||||
VoiceManifest(id="jf_nezumi", name="Nezumi", tags=("ja", "female")),
|
||||
VoiceManifest(id="jf_tebukuro", name="Tebukuro", tags=("ja", "female")),
|
||||
VoiceManifest(id="jm_kumo", name="Kumo", tags=("ja", "male")),
|
||||
VoiceManifest(id="pf_dora", name="Dora", tags=("pt", "female")),
|
||||
VoiceManifest(id="pm_alex", name="Alex", tags=("pt", "male")),
|
||||
VoiceManifest(id="pm_santa", name="Santa", tags=("pt", "male")),
|
||||
VoiceManifest(id="zf_xiaobei", name="Xiaobei", tags=("zh", "female")),
|
||||
VoiceManifest(id="zf_xiaoni", name="Xiaoni", tags=("zh", "female")),
|
||||
VoiceManifest(id="zf_xiaoxiao", name="Xiaoxiao", tags=("zh", "female")),
|
||||
VoiceManifest(id="zf_xiaoyi", name="Xiaoyi", tags=("zh", "female")),
|
||||
VoiceManifest(id="zm_yunjian", name="Yunjian", tags=("zh", "male")),
|
||||
VoiceManifest(id="zm_yunxi", name="Yunxi", tags=("zh", "male")),
|
||||
VoiceManifest(id="zm_yunxia", name="Yunxia", tags=("zh", "female")),
|
||||
VoiceManifest(id="zm_yunyang", name="Yunyang", tags=("zh", "male")),
|
||||
),
|
||||
)
|
||||
|
||||
MODEL_REQUIREMENTS: list[ModelManifest] = []
|
||||
|
||||
|
||||
def create_engine(
|
||||
context: HostContext,
|
||||
model_path: Path | None,
|
||||
config: EngineConfig,
|
||||
) -> Engine:
|
||||
"""Create a Kokoro engine instance.
|
||||
|
||||
This function is the plugin entry point. It must be atomic:
|
||||
succeed fully or raise EngineError and clean up.
|
||||
|
||||
Args:
|
||||
context: Host services (config dir, logger, http client).
|
||||
model_path: Resolved model path, or None for default.
|
||||
config: Engine initialization settings (device, etc.).
|
||||
|
||||
Returns:
|
||||
A fully initialized KokoroEngine instance.
|
||||
|
||||
Raises:
|
||||
EngineError: On failure. Cleans up partially created resources.
|
||||
"""
|
||||
try:
|
||||
KPipeline = _load_kpipeline()
|
||||
|
||||
# Determine repo_id from model_path or use default
|
||||
repo_id = "hexgrad/Kokoro-82M"
|
||||
if model_path is not None:
|
||||
# If a specific model path is provided, use it as repo_id
|
||||
repo_id = str(model_path)
|
||||
|
||||
pipeline = KPipeline(
|
||||
lang_code=config.lang_code,
|
||||
repo_id=repo_id,
|
||||
device=config.device,
|
||||
)
|
||||
|
||||
engine = KokoroEngine(pipeline)
|
||||
return engine
|
||||
except Exception as e:
|
||||
from abogen.tts_plugin.errors import EngineError as EngineErrorClass
|
||||
raise EngineErrorClass(f"Failed to create Kokoro engine: {e}") from e
|
||||
|
||||
+118
-118
@@ -1,118 +1,118 @@
|
||||
"""Kokoro Engine adapter for the TTS Plugin Architecture.
|
||||
|
||||
This module adapts the existing Kokoro backend to the new Engine/EngineSession
|
||||
protocol. It wraps the KokoroBackend without modifying it.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
from typing import Any
|
||||
|
||||
import numpy as np
|
||||
|
||||
from abogen.tts_plugin.capabilities import VoiceLister
|
||||
from abogen.tts_plugin.engine import Engine, EngineSession
|
||||
from abogen.tts_plugin.errors import EngineError
|
||||
from abogen.tts_plugin.manifest import VoiceManifest
|
||||
from abogen.tts_plugin.types import (
|
||||
AudioFormat,
|
||||
Duration,
|
||||
SynthesisRequest,
|
||||
SynthesizedAudio,
|
||||
)
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# Sample rate for Kokoro audio
|
||||
_KOKORO_SAMPLE_RATE = 24000
|
||||
|
||||
|
||||
class KokoroSession:
|
||||
"""EngineSession implementation for Kokoro.
|
||||
|
||||
Owns mutable execution state for synthesis.
|
||||
NOT thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def synthesize(self, request: SynthesisRequest) -> SynthesizedAudio:
|
||||
"""Synthesize audio from text using Kokoro."""
|
||||
if self._disposed:
|
||||
raise EngineError("Session disposed")
|
||||
|
||||
try:
|
||||
voice = request.voice.key
|
||||
speed = request.parameters.values.get("speed", 1.0)
|
||||
split_pattern = request.parameters.values.get("split_pattern", None)
|
||||
|
||||
audio_parts: list[np.ndarray] = []
|
||||
for segment in self._pipeline(
|
||||
request.text,
|
||||
voice=voice,
|
||||
speed=speed,
|
||||
split_pattern=split_pattern,
|
||||
):
|
||||
audio = segment.audio
|
||||
if hasattr(audio, "numpy"):
|
||||
audio = audio.numpy()
|
||||
audio_parts.append(np.asarray(audio, dtype="float32"))
|
||||
|
||||
if not audio_parts:
|
||||
return SynthesizedAudio(
|
||||
data=b"",
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=0.0),
|
||||
)
|
||||
|
||||
combined = np.concatenate(audio_parts).astype("float32", copy=False)
|
||||
audio_bytes = combined.tobytes()
|
||||
duration_seconds = len(combined) / _KOKORO_SAMPLE_RATE
|
||||
|
||||
return SynthesizedAudio(
|
||||
data=audio_bytes,
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=duration_seconds),
|
||||
)
|
||||
except EngineError:
|
||||
raise
|
||||
except Exception as e:
|
||||
raise EngineError(f"Synthesis failed: {e}") from e
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release session resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
|
||||
class KokoroEngine:
|
||||
"""Engine implementation for Kokoro.
|
||||
|
||||
Factory for KokoroSession instances. Stateless and thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def createSession(self) -> KokoroSession:
|
||||
"""Create a new KokoroSession."""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return KokoroSession(self._pipeline)
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release engine resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
def listVoices(self, sourceId: str) -> list[VoiceManifest]:
|
||||
"""List available Kokoro voices. Implements VoiceLister capability.
|
||||
|
||||
Note: Static voices are declared in the plugin manifest.
|
||||
This method is a fallback for dynamic plugins.
|
||||
"""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return []
|
||||
"""Kokoro Engine adapter for the TTS Plugin Architecture.
|
||||
|
||||
This module adapts the existing Kokoro backend to the new Engine/EngineSession
|
||||
protocol. It wraps the KokoroBackend without modifying it.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
from typing import Any
|
||||
|
||||
import numpy as np
|
||||
|
||||
from abogen.tts_plugin.capabilities import VoiceLister
|
||||
from abogen.tts_plugin.engine import Engine, EngineSession
|
||||
from abogen.tts_plugin.errors import EngineError
|
||||
from abogen.tts_plugin.manifest import VoiceManifest
|
||||
from abogen.tts_plugin.types import (
|
||||
AudioFormat,
|
||||
Duration,
|
||||
SynthesisRequest,
|
||||
SynthesizedAudio,
|
||||
)
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# Sample rate for Kokoro audio
|
||||
_KOKORO_SAMPLE_RATE = 24000
|
||||
|
||||
|
||||
class KokoroSession:
|
||||
"""EngineSession implementation for Kokoro.
|
||||
|
||||
Owns mutable execution state for synthesis.
|
||||
NOT thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def synthesize(self, request: SynthesisRequest) -> SynthesizedAudio:
|
||||
"""Synthesize audio from text using Kokoro."""
|
||||
if self._disposed:
|
||||
raise EngineError("Session disposed")
|
||||
|
||||
try:
|
||||
voice = request.voice.key
|
||||
speed = request.parameters.values.get("speed", 1.0)
|
||||
split_pattern = request.parameters.values.get("split_pattern", None)
|
||||
|
||||
audio_parts: list[np.ndarray] = []
|
||||
for segment in self._pipeline(
|
||||
request.text,
|
||||
voice=voice,
|
||||
speed=speed,
|
||||
split_pattern=split_pattern,
|
||||
):
|
||||
audio = segment.audio
|
||||
if hasattr(audio, "numpy"):
|
||||
audio = audio.numpy()
|
||||
audio_parts.append(np.asarray(audio, dtype="float32"))
|
||||
|
||||
if not audio_parts:
|
||||
return SynthesizedAudio(
|
||||
data=b"",
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=0.0),
|
||||
)
|
||||
|
||||
combined = np.concatenate(audio_parts).astype("float32", copy=False)
|
||||
audio_bytes = combined.tobytes()
|
||||
duration_seconds = len(combined) / _KOKORO_SAMPLE_RATE
|
||||
|
||||
return SynthesizedAudio(
|
||||
data=audio_bytes,
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=duration_seconds),
|
||||
)
|
||||
except EngineError:
|
||||
raise
|
||||
except Exception as e:
|
||||
raise EngineError(f"Synthesis failed: {e}") from e
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release session resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
|
||||
class KokoroEngine:
|
||||
"""Engine implementation for Kokoro.
|
||||
|
||||
Factory for KokoroSession instances. Stateless and thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def createSession(self) -> KokoroSession:
|
||||
"""Create a new KokoroSession."""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return KokoroSession(self._pipeline)
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release engine resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
def listVoices(self, sourceId: str) -> list[VoiceManifest]:
|
||||
"""List available Kokoro voices. Implements VoiceLister capability.
|
||||
|
||||
Note: Static voices are declared in the plugin manifest.
|
||||
This method is a fallback for dynamic plugins.
|
||||
"""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return []
|
||||
|
||||
+136
-136
@@ -1,136 +1,136 @@
|
||||
"""SuperTonic TTS Plugin for the TTS Plugin Architecture.
|
||||
|
||||
This plugin provides a SuperTonic-based TTS engine that implements the
|
||||
Plugin API contract. It wraps the existing SuperTonic backend in the
|
||||
new Engine/EngineSession architecture.
|
||||
|
||||
Exports:
|
||||
- PLUGIN_MANIFEST: PluginManifest
|
||||
- MODEL_REQUIREMENTS: list[ModelManifest]
|
||||
- create_engine: Factory function
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from abogen.tts_plugin.engine import Engine
|
||||
from abogen.tts_plugin.host_context import HostContext
|
||||
from abogen.tts_plugin.manifest import (
|
||||
AudioFormatManifest,
|
||||
EngineManifest,
|
||||
ModelManifest,
|
||||
ParameterManifest,
|
||||
PluginManifest,
|
||||
RequirementManifest,
|
||||
VoiceManifest,
|
||||
VoiceSourceManifest,
|
||||
)
|
||||
from abogen.tts_plugin.types import EngineConfig
|
||||
|
||||
from .engine import SuperTonicEngine
|
||||
|
||||
|
||||
def _load_supertonic_pipeline() -> Any:
|
||||
"""Lazy-load SuperTonic dependencies and create pipeline."""
|
||||
from plugins.supertonic.pipeline import SupertonicPipeline
|
||||
|
||||
return SupertonicPipeline(
|
||||
sample_rate=24000,
|
||||
auto_download=True,
|
||||
total_steps=5,
|
||||
)
|
||||
|
||||
|
||||
PLUGIN_MANIFEST = PluginManifest(
|
||||
id="supertonic",
|
||||
name="SuperTonic",
|
||||
version="0.1.0",
|
||||
api_version="1.0",
|
||||
description="SuperTonic TTS engine - fast high-quality text-to-speech",
|
||||
author="SuperTonic Team",
|
||||
capabilities=("voice_list",),
|
||||
requires=RequirementManifest(
|
||||
internet=False,
|
||||
),
|
||||
engine=EngineManifest(
|
||||
voiceSources=(
|
||||
VoiceSourceManifest(
|
||||
id="builtin",
|
||||
name="Built-in Voices",
|
||||
type="list",
|
||||
config={"voices": "See listVoices()"},
|
||||
),
|
||||
),
|
||||
parameters=(
|
||||
ParameterManifest(
|
||||
id="speed",
|
||||
name="Speed",
|
||||
description="Speech speed multiplier",
|
||||
type="float",
|
||||
default=1.0,
|
||||
min=0.7,
|
||||
max=2.0,
|
||||
step=0.1,
|
||||
),
|
||||
ParameterManifest(
|
||||
id="total_steps",
|
||||
name="Quality Steps",
|
||||
description="Inference steps (higher = better quality, slower)",
|
||||
type="int",
|
||||
default=5,
|
||||
min=2,
|
||||
max=15,
|
||||
step=1,
|
||||
),
|
||||
),
|
||||
audioFormats=(
|
||||
AudioFormatManifest(mime="audio/wav", extension="wav"),
|
||||
),
|
||||
),
|
||||
voices=(
|
||||
VoiceManifest(id="M1", name="Male 1", tags=("male",)),
|
||||
VoiceManifest(id="M2", name="Male 2", tags=("male",)),
|
||||
VoiceManifest(id="M3", name="Male 3", tags=("male",)),
|
||||
VoiceManifest(id="M4", name="Male 4", tags=("male",)),
|
||||
VoiceManifest(id="M5", name="Male 5", tags=("male",)),
|
||||
VoiceManifest(id="F1", name="Female 1", tags=("female",)),
|
||||
VoiceManifest(id="F2", name="Female 2", tags=("female",)),
|
||||
VoiceManifest(id="F3", name="Female 3", tags=("female",)),
|
||||
VoiceManifest(id="F4", name="Female 4", tags=("female",)),
|
||||
VoiceManifest(id="F5", name="Female 5", tags=("female",)),
|
||||
),
|
||||
)
|
||||
|
||||
MODEL_REQUIREMENTS: list[ModelManifest] = []
|
||||
|
||||
|
||||
def create_engine(
|
||||
context: HostContext,
|
||||
model_path: Path | None,
|
||||
config: EngineConfig,
|
||||
) -> Engine:
|
||||
"""Create a SuperTonic engine instance.
|
||||
|
||||
This function is the plugin entry point. It must be atomic:
|
||||
succeed fully or raise EngineError and clean up.
|
||||
|
||||
Args:
|
||||
context: Host services (config dir, logger, http client).
|
||||
model_path: Resolved model path, or None for default.
|
||||
config: Engine initialization settings (device, etc.).
|
||||
|
||||
Returns:
|
||||
A fully initialized SuperTonicEngine instance.
|
||||
|
||||
Raises:
|
||||
EngineError: On failure. Cleans up partially created resources.
|
||||
"""
|
||||
try:
|
||||
pipeline = _load_supertonic_pipeline()
|
||||
engine = SuperTonicEngine(pipeline)
|
||||
return engine
|
||||
except Exception as e:
|
||||
from abogen.tts_plugin.errors import EngineError as EngineErrorClass
|
||||
raise EngineErrorClass(f"Failed to create SuperTonic engine: {e}") from e
|
||||
"""SuperTonic TTS Plugin for the TTS Plugin Architecture.
|
||||
|
||||
This plugin provides a SuperTonic-based TTS engine that implements the
|
||||
Plugin API contract. It wraps the existing SuperTonic backend in the
|
||||
new Engine/EngineSession architecture.
|
||||
|
||||
Exports:
|
||||
- PLUGIN_MANIFEST: PluginManifest
|
||||
- MODEL_REQUIREMENTS: list[ModelManifest]
|
||||
- create_engine: Factory function
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from abogen.tts_plugin.engine import Engine
|
||||
from abogen.tts_plugin.host_context import HostContext
|
||||
from abogen.tts_plugin.manifest import (
|
||||
AudioFormatManifest,
|
||||
EngineManifest,
|
||||
ModelManifest,
|
||||
ParameterManifest,
|
||||
PluginManifest,
|
||||
RequirementManifest,
|
||||
VoiceManifest,
|
||||
VoiceSourceManifest,
|
||||
)
|
||||
from abogen.tts_plugin.types import EngineConfig
|
||||
|
||||
from .engine import SuperTonicEngine
|
||||
|
||||
|
||||
def _load_supertonic_pipeline() -> Any:
|
||||
"""Lazy-load SuperTonic dependencies and create pipeline."""
|
||||
from plugins.supertonic.pipeline import SupertonicPipeline
|
||||
|
||||
return SupertonicPipeline(
|
||||
sample_rate=24000,
|
||||
auto_download=True,
|
||||
total_steps=5,
|
||||
)
|
||||
|
||||
|
||||
PLUGIN_MANIFEST = PluginManifest(
|
||||
id="supertonic",
|
||||
name="SuperTonic",
|
||||
version="0.1.0",
|
||||
api_version="1.0",
|
||||
description="SuperTonic TTS engine - fast high-quality text-to-speech",
|
||||
author="SuperTonic Team",
|
||||
capabilities=("voice_list",),
|
||||
requires=RequirementManifest(
|
||||
internet=False,
|
||||
),
|
||||
engine=EngineManifest(
|
||||
voiceSources=(
|
||||
VoiceSourceManifest(
|
||||
id="builtin",
|
||||
name="Built-in Voices",
|
||||
type="list",
|
||||
config={"voices": "See listVoices()"},
|
||||
),
|
||||
),
|
||||
parameters=(
|
||||
ParameterManifest(
|
||||
id="speed",
|
||||
name="Speed",
|
||||
description="Speech speed multiplier",
|
||||
type="float",
|
||||
default=1.0,
|
||||
min=0.7,
|
||||
max=2.0,
|
||||
step=0.1,
|
||||
),
|
||||
ParameterManifest(
|
||||
id="total_steps",
|
||||
name="Quality Steps",
|
||||
description="Inference steps (higher = better quality, slower)",
|
||||
type="int",
|
||||
default=5,
|
||||
min=2,
|
||||
max=15,
|
||||
step=1,
|
||||
),
|
||||
),
|
||||
audioFormats=(
|
||||
AudioFormatManifest(mime="audio/wav", extension="wav"),
|
||||
),
|
||||
),
|
||||
voices=(
|
||||
VoiceManifest(id="M1", name="Male 1", tags=("male",)),
|
||||
VoiceManifest(id="M2", name="Male 2", tags=("male",)),
|
||||
VoiceManifest(id="M3", name="Male 3", tags=("male",)),
|
||||
VoiceManifest(id="M4", name="Male 4", tags=("male",)),
|
||||
VoiceManifest(id="M5", name="Male 5", tags=("male",)),
|
||||
VoiceManifest(id="F1", name="Female 1", tags=("female",)),
|
||||
VoiceManifest(id="F2", name="Female 2", tags=("female",)),
|
||||
VoiceManifest(id="F3", name="Female 3", tags=("female",)),
|
||||
VoiceManifest(id="F4", name="Female 4", tags=("female",)),
|
||||
VoiceManifest(id="F5", name="Female 5", tags=("female",)),
|
||||
),
|
||||
)
|
||||
|
||||
MODEL_REQUIREMENTS: list[ModelManifest] = []
|
||||
|
||||
|
||||
def create_engine(
|
||||
context: HostContext,
|
||||
model_path: Path | None,
|
||||
config: EngineConfig,
|
||||
) -> Engine:
|
||||
"""Create a SuperTonic engine instance.
|
||||
|
||||
This function is the plugin entry point. It must be atomic:
|
||||
succeed fully or raise EngineError and clean up.
|
||||
|
||||
Args:
|
||||
context: Host services (config dir, logger, http client).
|
||||
model_path: Resolved model path, or None for default.
|
||||
config: Engine initialization settings (device, etc.).
|
||||
|
||||
Returns:
|
||||
A fully initialized SuperTonicEngine instance.
|
||||
|
||||
Raises:
|
||||
EngineError: On failure. Cleans up partially created resources.
|
||||
"""
|
||||
try:
|
||||
pipeline = _load_supertonic_pipeline()
|
||||
engine = SuperTonicEngine(pipeline)
|
||||
return engine
|
||||
except Exception as e:
|
||||
from abogen.tts_plugin.errors import EngineError as EngineErrorClass
|
||||
raise EngineErrorClass(f"Failed to create SuperTonic engine: {e}") from e
|
||||
|
||||
+125
-125
@@ -1,125 +1,125 @@
|
||||
"""SuperTonic Engine adapter for the TTS Plugin Architecture.
|
||||
|
||||
This module adapts the existing SuperTonic backend to the new Engine/EngineSession
|
||||
protocol. It wraps the SupertonicPipeline without modifying it.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import io
|
||||
import logging
|
||||
from typing import Any
|
||||
|
||||
import numpy as np
|
||||
|
||||
from abogen.tts_plugin.capabilities import VoiceLister
|
||||
from abogen.tts_plugin.engine import Engine, EngineSession
|
||||
from abogen.tts_plugin.errors import EngineError
|
||||
from abogen.tts_plugin.manifest import VoiceManifest
|
||||
from abogen.tts_plugin.types import (
|
||||
AudioFormat,
|
||||
Duration,
|
||||
SynthesisRequest,
|
||||
SynthesizedAudio,
|
||||
)
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# Sample rate for SuperTonic audio
|
||||
_SUPERTONIC_SAMPLE_RATE = 24000
|
||||
|
||||
|
||||
class SuperTonicSession:
|
||||
"""EngineSession implementation for SuperTonic.
|
||||
|
||||
Owns mutable execution state for synthesis.
|
||||
NOT thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def synthesize(self, request: SynthesisRequest) -> SynthesizedAudio:
|
||||
"""Synthesize audio from text using SuperTonic."""
|
||||
if self._disposed:
|
||||
raise EngineError("Session disposed")
|
||||
|
||||
try:
|
||||
import soundfile as sf
|
||||
|
||||
voice = request.voice.key
|
||||
speed = float(request.parameters.values.get("speed", 1.0))
|
||||
total_steps = request.parameters.values.get("total_steps", None)
|
||||
split_pattern = request.parameters.values.get("split_pattern", None)
|
||||
|
||||
if total_steps is not None:
|
||||
total_steps = int(total_steps)
|
||||
|
||||
audio_parts: list[np.ndarray] = []
|
||||
for segment in self._pipeline(
|
||||
request.text,
|
||||
voice=voice,
|
||||
speed=speed,
|
||||
split_pattern=split_pattern,
|
||||
total_steps=total_steps,
|
||||
):
|
||||
audio_parts.append(segment.audio)
|
||||
|
||||
if not audio_parts:
|
||||
return SynthesizedAudio(
|
||||
data=b"",
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=0.0),
|
||||
)
|
||||
|
||||
combined = np.concatenate(audio_parts).astype("float32", copy=False)
|
||||
buf = io.BytesIO()
|
||||
sf.write(buf, combined, self._pipeline.sample_rate, format="WAV")
|
||||
audio_bytes = buf.getvalue()
|
||||
duration_seconds = len(combined) / self._pipeline.sample_rate
|
||||
|
||||
return SynthesizedAudio(
|
||||
data=audio_bytes,
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=duration_seconds),
|
||||
)
|
||||
except EngineError:
|
||||
raise
|
||||
except Exception as e:
|
||||
raise EngineError(f"Synthesis failed: {e}") from e
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release session resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
|
||||
class SuperTonicEngine:
|
||||
"""Engine implementation for SuperTonic.
|
||||
|
||||
Factory for SuperTonicSession instances. Stateless and thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def createSession(self) -> SuperTonicSession:
|
||||
"""Create a new SuperTonicSession."""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return SuperTonicSession(self._pipeline)
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release engine resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
def listVoices(self, sourceId: str) -> list[VoiceManifest]:
|
||||
"""List available SuperTonic voices. Implements VoiceLister capability.
|
||||
|
||||
Note: Static voice catalog is declared in plugin manifest.
|
||||
This method is retained for VoiceLister interface compliance.
|
||||
"""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return []
|
||||
"""SuperTonic Engine adapter for the TTS Plugin Architecture.
|
||||
|
||||
This module adapts the existing SuperTonic backend to the new Engine/EngineSession
|
||||
protocol. It wraps the SupertonicPipeline without modifying it.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import io
|
||||
import logging
|
||||
from typing import Any
|
||||
|
||||
import numpy as np
|
||||
|
||||
from abogen.tts_plugin.capabilities import VoiceLister
|
||||
from abogen.tts_plugin.engine import Engine, EngineSession
|
||||
from abogen.tts_plugin.errors import EngineError
|
||||
from abogen.tts_plugin.manifest import VoiceManifest
|
||||
from abogen.tts_plugin.types import (
|
||||
AudioFormat,
|
||||
Duration,
|
||||
SynthesisRequest,
|
||||
SynthesizedAudio,
|
||||
)
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# Sample rate for SuperTonic audio
|
||||
_SUPERTONIC_SAMPLE_RATE = 24000
|
||||
|
||||
|
||||
class SuperTonicSession:
|
||||
"""EngineSession implementation for SuperTonic.
|
||||
|
||||
Owns mutable execution state for synthesis.
|
||||
NOT thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def synthesize(self, request: SynthesisRequest) -> SynthesizedAudio:
|
||||
"""Synthesize audio from text using SuperTonic."""
|
||||
if self._disposed:
|
||||
raise EngineError("Session disposed")
|
||||
|
||||
try:
|
||||
import soundfile as sf
|
||||
|
||||
voice = request.voice.key
|
||||
speed = float(request.parameters.values.get("speed", 1.0))
|
||||
total_steps = request.parameters.values.get("total_steps", None)
|
||||
split_pattern = request.parameters.values.get("split_pattern", None)
|
||||
|
||||
if total_steps is not None:
|
||||
total_steps = int(total_steps)
|
||||
|
||||
audio_parts: list[np.ndarray] = []
|
||||
for segment in self._pipeline(
|
||||
request.text,
|
||||
voice=voice,
|
||||
speed=speed,
|
||||
split_pattern=split_pattern,
|
||||
total_steps=total_steps,
|
||||
):
|
||||
audio_parts.append(segment.audio)
|
||||
|
||||
if not audio_parts:
|
||||
return SynthesizedAudio(
|
||||
data=b"",
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=0.0),
|
||||
)
|
||||
|
||||
combined = np.concatenate(audio_parts).astype("float32", copy=False)
|
||||
buf = io.BytesIO()
|
||||
sf.write(buf, combined, self._pipeline.sample_rate, format="WAV")
|
||||
audio_bytes = buf.getvalue()
|
||||
duration_seconds = len(combined) / self._pipeline.sample_rate
|
||||
|
||||
return SynthesizedAudio(
|
||||
data=audio_bytes,
|
||||
format=AudioFormat(mime="audio/wav", extension="wav"),
|
||||
duration=Duration(seconds=duration_seconds),
|
||||
)
|
||||
except EngineError:
|
||||
raise
|
||||
except Exception as e:
|
||||
raise EngineError(f"Synthesis failed: {e}") from e
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release session resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
|
||||
class SuperTonicEngine:
|
||||
"""Engine implementation for SuperTonic.
|
||||
|
||||
Factory for SuperTonicSession instances. Stateless and thread-safe.
|
||||
"""
|
||||
|
||||
def __init__(self, pipeline: Any) -> None:
|
||||
self._pipeline = pipeline
|
||||
self._disposed = False
|
||||
|
||||
def createSession(self) -> SuperTonicSession:
|
||||
"""Create a new SuperTonicSession."""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return SuperTonicSession(self._pipeline)
|
||||
|
||||
def dispose(self) -> None:
|
||||
"""Release engine resources. Idempotent."""
|
||||
self._disposed = True
|
||||
|
||||
def listVoices(self, sourceId: str) -> list[VoiceManifest]:
|
||||
"""List available SuperTonic voices. Implements VoiceLister capability.
|
||||
|
||||
Note: Static voice catalog is declared in plugin manifest.
|
||||
This method is retained for VoiceLister interface compliance.
|
||||
"""
|
||||
if self._disposed:
|
||||
raise EngineError("Engine disposed")
|
||||
return []
|
||||
|
||||
+266
-266
@@ -1,266 +1,266 @@
|
||||
"""SuperTonic Pipeline — self-contained TTS pipeline for the plugin.
|
||||
|
||||
This module provides the SuperTonicPipeline class and supporting utilities
|
||||
used by the SuperTonic plugin. It is independent of the legacy
|
||||
abogen.tts_backends module.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import ast
|
||||
import logging
|
||||
import re
|
||||
from typing import Any, Iterable, Iterator, Optional
|
||||
|
||||
import numpy as np
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
def _ensure_float32_mono(wav: Any) -> np.ndarray:
|
||||
arr = np.asarray(wav, dtype="float32")
|
||||
if arr.ndim == 2:
|
||||
if arr.shape[0] == 1 and arr.shape[1] > 1:
|
||||
arr = arr.reshape(-1)
|
||||
else:
|
||||
arr = arr[:, 0]
|
||||
return arr.reshape(-1)
|
||||
|
||||
|
||||
def _resample_linear(audio: np.ndarray, src_rate: int, dst_rate: int) -> np.ndarray:
|
||||
if src_rate == dst_rate:
|
||||
return audio
|
||||
if audio.size == 0:
|
||||
return audio
|
||||
ratio = dst_rate / float(src_rate)
|
||||
new_len = int(round(audio.size * ratio))
|
||||
if new_len <= 1:
|
||||
return np.zeros(0, dtype="float32")
|
||||
x_old = np.linspace(0.0, 1.0, num=audio.size, endpoint=False)
|
||||
x_new = np.linspace(0.0, 1.0, num=new_len, endpoint=False)
|
||||
return np.interp(x_new, x_old, audio).astype("float32", copy=False)
|
||||
|
||||
|
||||
def _split_text(
|
||||
text: str, *, split_pattern: Optional[str], max_chunk_length: int
|
||||
) -> list[str]:
|
||||
stripped = (text or "").strip()
|
||||
if not stripped:
|
||||
return []
|
||||
parts: list[str]
|
||||
if split_pattern:
|
||||
try:
|
||||
parts = [p.strip() for p in re.split(split_pattern, stripped) if p.strip()]
|
||||
except re.error:
|
||||
parts = [stripped]
|
||||
else:
|
||||
parts = [stripped]
|
||||
|
||||
result: list[str] = []
|
||||
for part in parts:
|
||||
if len(part) <= max_chunk_length:
|
||||
result.append(part)
|
||||
continue
|
||||
start = 0
|
||||
while start < len(part):
|
||||
end = min(len(part), start + max_chunk_length)
|
||||
if end < len(part):
|
||||
ws = part.rfind(" ", start, end)
|
||||
if ws > start + 40:
|
||||
end = ws
|
||||
chunk = part[start:end].strip()
|
||||
if chunk:
|
||||
result.append(chunk)
|
||||
start = end
|
||||
return result
|
||||
|
||||
|
||||
_UNSUPPORTED_CHARS_RE = re.compile(
|
||||
r"unsupported character\(s\):\s*(\[[^\]]*\])", re.IGNORECASE
|
||||
)
|
||||
|
||||
|
||||
def _parse_unsupported_characters(error: BaseException) -> list[str]:
|
||||
"""Best-effort extraction of unsupported characters from SuperTonic errors."""
|
||||
message = " ".join(
|
||||
str(part) for part in getattr(error, "args", ()) if part is not None
|
||||
) or str(error)
|
||||
match = _UNSUPPORTED_CHARS_RE.search(message)
|
||||
if not match:
|
||||
return []
|
||||
|
||||
raw = match.group(1)
|
||||
try:
|
||||
value = ast.literal_eval(raw)
|
||||
except Exception:
|
||||
return []
|
||||
|
||||
if isinstance(value, (list, tuple)):
|
||||
out: list[str] = []
|
||||
for item in value:
|
||||
if item is None:
|
||||
continue
|
||||
s = str(item)
|
||||
if s:
|
||||
out.append(s)
|
||||
return out
|
||||
|
||||
if isinstance(value, str) and value:
|
||||
return [value]
|
||||
|
||||
return []
|
||||
|
||||
|
||||
def _remove_unsupported_characters(text: str, unsupported: Iterable[str]) -> str:
|
||||
result = text
|
||||
for item in unsupported:
|
||||
if not item:
|
||||
continue
|
||||
result = result.replace(item, "")
|
||||
return result
|
||||
|
||||
|
||||
def _configure_supertonic_gpu() -> None:
|
||||
"""Patch supertonic's config to enable GPU acceleration if available."""
|
||||
try:
|
||||
import onnxruntime as ort
|
||||
|
||||
available = ort.get_available_providers()
|
||||
|
||||
providers = []
|
||||
if "CUDAExecutionProvider" in available:
|
||||
providers.append("CUDAExecutionProvider")
|
||||
providers.append("CPUExecutionProvider")
|
||||
|
||||
import supertonic.config as supertonic_config
|
||||
import supertonic.loader as supertonic_loader
|
||||
|
||||
supertonic_config.DEFAULT_ONNX_PROVIDERS = providers
|
||||
supertonic_loader.DEFAULT_ONNX_PROVIDERS = providers
|
||||
logger.info("Supertonic ONNX providers configured: %s", providers)
|
||||
except Exception as exc:
|
||||
logger.warning("Could not configure supertonic GPU providers: %s", exc)
|
||||
|
||||
|
||||
class SupertonicSegment:
|
||||
"""A single synthesized audio segment."""
|
||||
|
||||
__slots__ = ("graphemes", "audio")
|
||||
|
||||
def __init__(self, graphemes: str, audio: np.ndarray) -> None:
|
||||
self.graphemes = graphemes
|
||||
self.audio = audio
|
||||
|
||||
|
||||
class SupertonicPipeline:
|
||||
"""Minimal adapter that mimics Kokoro's pipeline iteration interface."""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
sample_rate: int,
|
||||
auto_download: bool = True,
|
||||
total_steps: int = 5,
|
||||
max_chunk_length: int = 300,
|
||||
) -> None:
|
||||
self.sample_rate = int(sample_rate)
|
||||
self.total_steps = int(total_steps)
|
||||
self.max_chunk_length = int(max_chunk_length)
|
||||
|
||||
_configure_supertonic_gpu()
|
||||
|
||||
try:
|
||||
from supertonic import TTS # type: ignore[import-not-found]
|
||||
except Exception as exc: # pragma: no cover
|
||||
raise RuntimeError(
|
||||
"Supertonic is not installed. Install it with `pip install supertonic`."
|
||||
) from exc
|
||||
|
||||
self._tts = TTS(auto_download=auto_download)
|
||||
|
||||
def __call__(
|
||||
self,
|
||||
text: str,
|
||||
*,
|
||||
voice: str,
|
||||
speed: float,
|
||||
split_pattern: Optional[str] = None,
|
||||
total_steps: Optional[int] = None,
|
||||
) -> Iterator[SupertonicSegment]:
|
||||
voice_name = (voice or "").strip() or "M1"
|
||||
steps = int(total_steps) if total_steps is not None else self.total_steps
|
||||
steps = max(2, min(15, steps))
|
||||
speed_value = float(speed) if speed is not None else 1.0
|
||||
speed_value = max(0.7, min(2.0, speed_value))
|
||||
|
||||
style = self._tts.get_voice_style(voice_name=voice_name)
|
||||
chunks = _split_text(
|
||||
text, split_pattern=split_pattern, max_chunk_length=self.max_chunk_length
|
||||
)
|
||||
for chunk in chunks:
|
||||
chunk_to_speak = chunk
|
||||
removed: set[str] = set()
|
||||
last_exc: Exception | None = None
|
||||
|
||||
for attempt in range(3):
|
||||
try:
|
||||
wav, duration = self._tts.synthesize(
|
||||
text=chunk_to_speak,
|
||||
voice_style=style,
|
||||
total_steps=steps,
|
||||
speed=speed_value,
|
||||
max_chunk_length=self.max_chunk_length,
|
||||
silence_duration=0.0,
|
||||
verbose=False,
|
||||
)
|
||||
break
|
||||
except ValueError as exc:
|
||||
last_exc = exc
|
||||
unsupported = _parse_unsupported_characters(exc)
|
||||
if not unsupported:
|
||||
raise
|
||||
|
||||
removed.update(unsupported)
|
||||
sanitized = _remove_unsupported_characters(
|
||||
chunk_to_speak, unsupported
|
||||
).strip()
|
||||
|
||||
if sanitized == chunk_to_speak.strip():
|
||||
raise
|
||||
|
||||
chunk_to_speak = sanitized
|
||||
if not chunk_to_speak:
|
||||
logger.warning(
|
||||
"SuperTonic: dropped a chunk after removing unsupported characters: %s",
|
||||
sorted(removed),
|
||||
)
|
||||
break
|
||||
|
||||
if attempt == 0:
|
||||
logger.warning(
|
||||
"SuperTonic: removed unsupported characters %s and retried.",
|
||||
sorted(removed),
|
||||
)
|
||||
else:
|
||||
assert last_exc is not None
|
||||
raise last_exc
|
||||
|
||||
if not chunk_to_speak:
|
||||
continue
|
||||
|
||||
audio = _ensure_float32_mono(wav)
|
||||
|
||||
src_rate = self.sample_rate
|
||||
try:
|
||||
dur = float(duration)
|
||||
if dur > 0 and audio.size > 0:
|
||||
inferred = int(round(audio.size / dur))
|
||||
if 8000 <= inferred <= 96000:
|
||||
src_rate = inferred
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if src_rate != self.sample_rate:
|
||||
audio = _resample_linear(audio, src_rate, self.sample_rate)
|
||||
|
||||
yield SupertonicSegment(graphemes=chunk_to_speak, audio=audio)
|
||||
"""SuperTonic Pipeline — self-contained TTS pipeline for the plugin.
|
||||
|
||||
This module provides the SuperTonicPipeline class and supporting utilities
|
||||
used by the SuperTonic plugin. It is independent of the legacy
|
||||
abogen.tts_backends module.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import ast
|
||||
import logging
|
||||
import re
|
||||
from typing import Any, Iterable, Iterator, Optional
|
||||
|
||||
import numpy as np
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
def _ensure_float32_mono(wav: Any) -> np.ndarray:
|
||||
arr = np.asarray(wav, dtype="float32")
|
||||
if arr.ndim == 2:
|
||||
if arr.shape[0] == 1 and arr.shape[1] > 1:
|
||||
arr = arr.reshape(-1)
|
||||
else:
|
||||
arr = arr[:, 0]
|
||||
return arr.reshape(-1)
|
||||
|
||||
|
||||
def _resample_linear(audio: np.ndarray, src_rate: int, dst_rate: int) -> np.ndarray:
|
||||
if src_rate == dst_rate:
|
||||
return audio
|
||||
if audio.size == 0:
|
||||
return audio
|
||||
ratio = dst_rate / float(src_rate)
|
||||
new_len = int(round(audio.size * ratio))
|
||||
if new_len <= 1:
|
||||
return np.zeros(0, dtype="float32")
|
||||
x_old = np.linspace(0.0, 1.0, num=audio.size, endpoint=False)
|
||||
x_new = np.linspace(0.0, 1.0, num=new_len, endpoint=False)
|
||||
return np.interp(x_new, x_old, audio).astype("float32", copy=False)
|
||||
|
||||
|
||||
def _split_text(
|
||||
text: str, *, split_pattern: Optional[str], max_chunk_length: int
|
||||
) -> list[str]:
|
||||
stripped = (text or "").strip()
|
||||
if not stripped:
|
||||
return []
|
||||
parts: list[str]
|
||||
if split_pattern:
|
||||
try:
|
||||
parts = [p.strip() for p in re.split(split_pattern, stripped) if p.strip()]
|
||||
except re.error:
|
||||
parts = [stripped]
|
||||
else:
|
||||
parts = [stripped]
|
||||
|
||||
result: list[str] = []
|
||||
for part in parts:
|
||||
if len(part) <= max_chunk_length:
|
||||
result.append(part)
|
||||
continue
|
||||
start = 0
|
||||
while start < len(part):
|
||||
end = min(len(part), start + max_chunk_length)
|
||||
if end < len(part):
|
||||
ws = part.rfind(" ", start, end)
|
||||
if ws > start + 40:
|
||||
end = ws
|
||||
chunk = part[start:end].strip()
|
||||
if chunk:
|
||||
result.append(chunk)
|
||||
start = end
|
||||
return result
|
||||
|
||||
|
||||
_UNSUPPORTED_CHARS_RE = re.compile(
|
||||
r"unsupported character\(s\):\s*(\[[^\]]*\])", re.IGNORECASE
|
||||
)
|
||||
|
||||
|
||||
def _parse_unsupported_characters(error: BaseException) -> list[str]:
|
||||
"""Best-effort extraction of unsupported characters from SuperTonic errors."""
|
||||
message = " ".join(
|
||||
str(part) for part in getattr(error, "args", ()) if part is not None
|
||||
) or str(error)
|
||||
match = _UNSUPPORTED_CHARS_RE.search(message)
|
||||
if not match:
|
||||
return []
|
||||
|
||||
raw = match.group(1)
|
||||
try:
|
||||
value = ast.literal_eval(raw)
|
||||
except Exception:
|
||||
return []
|
||||
|
||||
if isinstance(value, (list, tuple)):
|
||||
out: list[str] = []
|
||||
for item in value:
|
||||
if item is None:
|
||||
continue
|
||||
s = str(item)
|
||||
if s:
|
||||
out.append(s)
|
||||
return out
|
||||
|
||||
if isinstance(value, str) and value:
|
||||
return [value]
|
||||
|
||||
return []
|
||||
|
||||
|
||||
def _remove_unsupported_characters(text: str, unsupported: Iterable[str]) -> str:
|
||||
result = text
|
||||
for item in unsupported:
|
||||
if not item:
|
||||
continue
|
||||
result = result.replace(item, "")
|
||||
return result
|
||||
|
||||
|
||||
def _configure_supertonic_gpu() -> None:
|
||||
"""Patch supertonic's config to enable GPU acceleration if available."""
|
||||
try:
|
||||
import onnxruntime as ort
|
||||
|
||||
available = ort.get_available_providers()
|
||||
|
||||
providers = []
|
||||
if "CUDAExecutionProvider" in available:
|
||||
providers.append("CUDAExecutionProvider")
|
||||
providers.append("CPUExecutionProvider")
|
||||
|
||||
import supertonic.config as supertonic_config
|
||||
import supertonic.loader as supertonic_loader
|
||||
|
||||
supertonic_config.DEFAULT_ONNX_PROVIDERS = providers
|
||||
supertonic_loader.DEFAULT_ONNX_PROVIDERS = providers
|
||||
logger.info("Supertonic ONNX providers configured: %s", providers)
|
||||
except Exception as exc:
|
||||
logger.warning("Could not configure supertonic GPU providers: %s", exc)
|
||||
|
||||
|
||||
class SupertonicSegment:
|
||||
"""A single synthesized audio segment."""
|
||||
|
||||
__slots__ = ("graphemes", "audio")
|
||||
|
||||
def __init__(self, graphemes: str, audio: np.ndarray) -> None:
|
||||
self.graphemes = graphemes
|
||||
self.audio = audio
|
||||
|
||||
|
||||
class SupertonicPipeline:
|
||||
"""Minimal adapter that mimics Kokoro's pipeline iteration interface."""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
sample_rate: int,
|
||||
auto_download: bool = True,
|
||||
total_steps: int = 5,
|
||||
max_chunk_length: int = 300,
|
||||
) -> None:
|
||||
self.sample_rate = int(sample_rate)
|
||||
self.total_steps = int(total_steps)
|
||||
self.max_chunk_length = int(max_chunk_length)
|
||||
|
||||
_configure_supertonic_gpu()
|
||||
|
||||
try:
|
||||
from supertonic import TTS # type: ignore[import-not-found]
|
||||
except Exception as exc: # pragma: no cover
|
||||
raise RuntimeError(
|
||||
"Supertonic is not installed. Install it with `pip install supertonic`."
|
||||
) from exc
|
||||
|
||||
self._tts = TTS(auto_download=auto_download)
|
||||
|
||||
def __call__(
|
||||
self,
|
||||
text: str,
|
||||
*,
|
||||
voice: str,
|
||||
speed: float,
|
||||
split_pattern: Optional[str] = None,
|
||||
total_steps: Optional[int] = None,
|
||||
) -> Iterator[SupertonicSegment]:
|
||||
voice_name = (voice or "").strip() or "M1"
|
||||
steps = int(total_steps) if total_steps is not None else self.total_steps
|
||||
steps = max(2, min(15, steps))
|
||||
speed_value = float(speed) if speed is not None else 1.0
|
||||
speed_value = max(0.7, min(2.0, speed_value))
|
||||
|
||||
style = self._tts.get_voice_style(voice_name=voice_name)
|
||||
chunks = _split_text(
|
||||
text, split_pattern=split_pattern, max_chunk_length=self.max_chunk_length
|
||||
)
|
||||
for chunk in chunks:
|
||||
chunk_to_speak = chunk
|
||||
removed: set[str] = set()
|
||||
last_exc: Exception | None = None
|
||||
|
||||
for attempt in range(3):
|
||||
try:
|
||||
wav, duration = self._tts.synthesize(
|
||||
text=chunk_to_speak,
|
||||
voice_style=style,
|
||||
total_steps=steps,
|
||||
speed=speed_value,
|
||||
max_chunk_length=self.max_chunk_length,
|
||||
silence_duration=0.0,
|
||||
verbose=False,
|
||||
)
|
||||
break
|
||||
except ValueError as exc:
|
||||
last_exc = exc
|
||||
unsupported = _parse_unsupported_characters(exc)
|
||||
if not unsupported:
|
||||
raise
|
||||
|
||||
removed.update(unsupported)
|
||||
sanitized = _remove_unsupported_characters(
|
||||
chunk_to_speak, unsupported
|
||||
).strip()
|
||||
|
||||
if sanitized == chunk_to_speak.strip():
|
||||
raise
|
||||
|
||||
chunk_to_speak = sanitized
|
||||
if not chunk_to_speak:
|
||||
logger.warning(
|
||||
"SuperTonic: dropped a chunk after removing unsupported characters: %s",
|
||||
sorted(removed),
|
||||
)
|
||||
break
|
||||
|
||||
if attempt == 0:
|
||||
logger.warning(
|
||||
"SuperTonic: removed unsupported characters %s and retried.",
|
||||
sorted(removed),
|
||||
)
|
||||
else:
|
||||
assert last_exc is not None
|
||||
raise last_exc
|
||||
|
||||
if not chunk_to_speak:
|
||||
continue
|
||||
|
||||
audio = _ensure_float32_mono(wav)
|
||||
|
||||
src_rate = self.sample_rate
|
||||
try:
|
||||
dur = float(duration)
|
||||
if dur > 0 and audio.size > 0:
|
||||
inferred = int(round(audio.size / dur))
|
||||
if 8000 <= inferred <= 96000:
|
||||
src_rate = inferred
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if src_rate != self.sample_rate:
|
||||
audio = _resample_linear(audio, src_rate, self.sample_rate)
|
||||
|
||||
yield SupertonicSegment(graphemes=chunk_to_speak, audio=audio)
|
||||
|
||||
Reference in New Issue
Block a user