livekit-plugins-flashtts 1.1.0__py3-none-any.whl

livekit/plugins/flashtts/__init__.py

@@ -0,0 +1,25 @@
+from .tts import TTS
+from .version import __version__
+
+__all__ = ["TTS", "__version__"]
+
+from livekit.agents import Plugin
+
+from .log import logger
+
+
+class FlashTTSPlugin(Plugin):
+    def __init__(self):
+        super().__init__(__name__, __version__, __package__, logger)
+
+
+Plugin.register_plugin(FlashTTSPlugin())
+
+# Cleanup docs of unexported modules
+_module = dir()
+NOT_IN_ALL = [m for m in _module if m not in __all__]
+
+__pdoc__ = {}
+
+for n in NOT_IN_ALL:
+    __pdoc__[n] = False

livekit/plugins/flashtts/log.py

@@ -0,0 +1,3 @@
+from logging import getLogger
+
+logger = getLogger("livekit.plugins.flashtts")

livekit/plugins/flashtts/tts.py

@@ -0,0 +1,246 @@
+from __future__ import annotations
+
+import time
+from typing import Dict, Literal, Optional
+import os
+
+import aiohttp
+from pydantic import BaseModel, Field
+from osc_data.text_stream import TextStreamSentencizer
+
+from livekit.agents import (
+    APIConnectOptions,
+    tts,
+    utils,
+)
+from livekit.agents.types import DEFAULT_API_CONNECT_OPTIONS
+
+from .log import logger
+
+
+class TTSOptions(BaseModel):
+    base_url: str | None = None
+    api_key: str | None = None
+    sample_rate: int = 16000
+    name: Optional[str] = Field(
+        default=None,
+        description="The name of the voice character to be used for speech synthesis.",
+    )
+    pitch: Optional[Literal["very_low", "low", "moderate", "high", "very_high"]] = (
+        Field(
+            default=None,
+            description="Specifies the pitch level for the generated audio. Valid options: 'very_low', 'low', 'moderate', 'high', 'very_high'.",
+        )
+    )
+    speed: Optional[Literal["very_low", "low", "moderate", "high", "very_high"]] = (
+        Field(
+            default=None,
+            description="Specifies the speed level of the audio output. Valid options: 'very_low', 'low', 'moderate', 'high', 'very_high'.",
+        )
+    )
+    temperature: float = Field(
+        default=0.9,
+        description="Controls the randomness of the speech synthesis. A higher temperature produces more diverse outputs.",
+    )
+    top_k: int = Field(
+        default=50,
+        description="Limits the sampling to the top 'k' most probable tokens during generation.",
+    )
+    top_p: float = Field(
+        default=0.95,
+        description="Nucleus sampling threshold: only tokens with a cumulative probability up to 'top_p' are considered.",
+    )
+    repetition_penalty: float = Field(
+        default=1.0,
+        description="Controls the repetition penalty applied to the generated text. "
+        "Higher values penalize repeated words and phrases.",
+    )
+    max_tokens: int = Field(
+        default=32768,
+        description="Specifies the maximum number of tokens to generate in the output.",
+    )
+    length_threshold: int = Field(
+        default=1000000,
+        description="If the input text exceeds this token length threshold, it will be split into multiple segments for synthesis.",
+    )
+    window_size: int = Field(
+        default=100000,
+        description="Determines the window size for each text segment when performing segmentation on longer texts.",
+    )
+    stream: bool = Field(
+        default=True,
+        description="Indicates whether the audio output should be streamed in real-time (True) or returned only after complete synthesis (False).",
+    )
+    response_format: Literal["mp3", "opus", "aac", "flac", "wav", "pcm"] = Field(
+        default="pcm",
+        description=(
+            "The format in which to return audio. Supported formats: mp3, opus, aac, flac, wav, pcm. "
+            "Note: PCM returns raw 16-bit samples without headers and AAC is not currently supported."
+        ),
+    )
+
+    def get_http_url(self) -> str:
+        return f"{self.base_url}/speak"
+
+    def get_http_headers(self) -> Dict:
+        return {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {self.api_key}",
+        }
+
+    def get_query_params(self, text: str) -> Dict:
+        if self.api_key is None:
+            self.api_key = os.environ.get("FLASHTTS_API_KEY", None)
+        if self.base_url is None:
+            self.base_url = os.environ.get("FLASHTTS_BASE_URL", "http://localhost:8000")
+        params = self.model_dump()
+        params["text"] = text
+        return params
+
+
+class TTS(tts.TTS):
+    def __init__(
+        self,
+        base_url: str | None = None,
+        api_key: str | None = None,
+        sample_rate: int = 16000,
+        voice: Optional[str] = "female",
+        pitch: Optional[
+            Literal["very_low", "low", "moderate", "high", "very_high"]
+        ] = None,
+        speed: Optional[
+            Literal["very_low", "low", "moderate", "high", "very_high"]
+        ] = None,
+        temperature: float = 0.9,
+        top_k: int = 50,
+        top_p: float = 0.95,
+        repetition_penalty: float = 1.0,
+        max_tokens: int = 32768,
+        http_session: aiohttp.ClientSession | None = None,
+        max_session_duration: float = 600,
+    ):
+        """flashtts
+
+        Args:
+            base_url (str | None, optional): Base URL. Defaults to None.
+            api_key (str | None, optional): API key. Defaults to None.
+            sample_rate (int, optional): Sample rate. Defaults to 16000.
+            voice (Optional[str], optional): voice name. Defaults to "female".
+            pitch (Optional[Literal[ "very_low", "low", "moderate", "high", "very_high" ]], optional): Pitch. Defaults to None.
+            speed (Optional[Literal[ "very_low", "low", "moderate", "high", "very_high" ]], optional): Speed. Defaults to None.
+            temperature (float, optional): Temperature. Defaults to 0.9.
+            top_k (int, optional): Top k. Defaults to 50.
+            top_p (float, optional): Top p. Defaults to 0.95.
+            repetition_penalty (float, optional): Repetition penalty. Defaults to 1.0.
+            max_tokens (int, optional): Max tokens. Defaults to 4096.
+            stream (bool, optional): Stream. Defaults to False.
+            http_session (aiohttp.ClientSession | None, optional): HTTP session. Defaults to None.
+            max_session_duration (float, optional): Max session duration. Defaults to 600.
+        """
+        super().__init__(
+            capabilities=tts.TTSCapabilities(streaming=True),
+            sample_rate=sample_rate,
+            num_channels=1,
+        )
+        self._opts = TTSOptions(
+            base_url=base_url,
+            api_key=api_key,
+            sample_rate=sample_rate,
+            name=voice,
+            pitch=pitch,
+            speed=speed,
+            temperature=temperature,
+            top_k=top_k,
+            top_p=top_p,
+            repetition_penalty=repetition_penalty,
+            max_tokens=max_tokens,
+        )
+        self._session = http_session
+
+    def _ensure_session(self) -> aiohttp.ClientSession:
+        if self._session is None:
+            self._session = utils.http_context.http_session()
+
+        return self._session
+
+    def synthesize(
+        self, text, *, conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS
+    ):
+        raise NotImplementedError("Minimax TTS does not support synthesize method")
+
+    def stream(self, *, conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS):
+        return SynthesizeStream(
+            tts=self,
+            conn_options=conn_options,
+            opts=self._opts,
+            session=self._ensure_session(),
+        )
+
+
+class SynthesizeStream(tts.SynthesizeStream):
+    def __init__(
+        self,
+        *,
+        tts: TTS,
+        opts: TTSOptions,
+        session: aiohttp.ClientSession,
+        conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS,
+    ):
+        super().__init__(tts=tts, conn_options=conn_options)
+        self._opts, self._session = opts, session
+
+    async def _run(self) -> None:
+        request_id = utils.shortuuid()
+        audio_bstream = utils.audio.AudioByteStream(
+            sample_rate=self._opts.sample_rate,
+            num_channels=1,
+        )
+        emitter = tts.SynthesizedAudioEmitter(
+            event_ch=self._event_ch,
+            request_id=request_id,
+        )
+        splitter = TextStreamSentencizer()
+        first_sentence_spend = None
+        start_time = time.perf_counter()
+        async for token in self._input_ch:
+            if isinstance(token, self._FlushSentinel):
+                sentences = splitter.flush()
+
+            else:
+                sentences = splitter.push(text=token)
+            for sentence in sentences:
+                if first_sentence_spend is None:
+                    first_sentence_spend = time.perf_counter() - start_time
+                    logger.info(
+                        "llm first sentence",
+                        extra={"spent": str(first_sentence_spend)},
+                    )
+                if len(sentence.strip()) > 0:
+                    first_response_spend = None
+                    logger.info("tts start", extra={"sentence": sentence})
+                    data = self._opts.get_query_params(text=sentence)
+                    if first_response_spend is None:
+                        start_time = time.perf_counter()
+                    async with self._session.post(
+                        self._opts.get_http_url(),
+                        json=data,
+                        timeout=aiohttp.ClientTimeout(
+                            total=30,
+                            sock_connect=self._conn_options.timeout,
+                        ),
+                        headers=self._opts.get_http_headers(),
+                    ) as resp:
+                        resp.raise_for_status()
+                        async for data in resp.content:
+                            if first_response_spend is None:
+                                first_response_spend = time.perf_counter() - start_time
+                                logger.info(
+                                    "tts first response",
+                                    extra={"spent": str(first_response_spend)},
+                                )
+                            for frame in audio_bstream.write(data=data):
+                                emitter.push(frame)
+                        for frame in audio_bstream.flush():
+                            emitter.push(frame)
+                    emitter.flush()
+                    logger.info("tts end")

livekit/plugins/flashtts/version.py

@@ -0,0 +1 @@
+__version__ = "1.1.0"

livekit_plugins_flashtts-1.1.0.dist-info/METADATA

@@ -0,0 +1,64 @@
+Metadata-Version: 2.4
+Name: livekit-plugins-flashtts
+Version: 1.1.0
+Summary: LiveKit Agent Plugins for FlashTTS
+Author-email: wangmengdi <790990241@qq.com>
+Keywords: audio,livekit,realtime,video,webrtc
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Topic :: Multimedia :: Sound/Audio
+Classifier: Topic :: Multimedia :: Video
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Requires-Dist: livekit-agents~=1.0.23
+Requires-Dist: osc-data>=0.1.7.post0
+Requires-Dist: pydantic
+Description-Content-Type: text/markdown
+
+# 简介
+
+[FlashTTS](https://github.com/HuiResearch/FlashTTS)是一个开源的TTS推理框架，你可以使用它部署Spark-TTS，MegaTTS等开源模型。
+
+## 安装
+```python
+pip install livekit-plugins-flashtts
+```
+
+## 环境变量
+
+- `FLASHTTS_API_URL`，FlashTTS的API地址，默认值为`http://localhost:8000`。
+- `FLASHTTS_API_KEY`，FlashTTS的API密钥，默认值为空。
+
+## 使用
+
+
+以下是一个使用FlashTTS插件的示例：
+
+```python
+from livekit.agents import Agent, AgentSession, JobContext, cli, WorkerOptions
+from livekit.plugins import flashtts
+from dotenv import load_dotenv
+
+
+async def entry_point(ctx: JobContext):
+    
+    await ctx.connect()
+    
+    agent = Agent(instructions="You are a helpful assistant.")
+
+    session = AgentSession(
+        tts=flashtts.TTS(voice="female"),
+    )
+    
+    await session.start(agent=agent, room=ctx.room)
+
+
+if __name__ == "__main__":
+    load_dotenv()
+    cli.run_app(WorkerOptions(entrypoint_fnc=entry_point))
+```
+

livekit_plugins_flashtts-1.1.0.dist-info/RECORD

@@ -0,0 +1,8 @@
+livekit/plugins/flashtts/__init__.py,sha256=E6bq3pwf5Gon4jhZy1xw2OhYe9WOBzA-ZDdjwtaUfkk,485
+livekit/plugins/flashtts/log.py,sha256=-C-i1PZKWxlwxL5voj6sPz9qLVuWOh0V8CKEdKLGTZc,78
+livekit/plugins/flashtts/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+livekit/plugins/flashtts/tts.py,sha256=LocuED2GlK55TV9JNI1hhVlvEerwIcvl_gBfgCRgzyA,9686
+livekit/plugins/flashtts/version.py,sha256=LGVQyDsWifdACo7qztwb8RWWHds1E7uQ-ZqD8SAjyw4,22
+livekit_plugins_flashtts-1.1.0.dist-info/METADATA,sha256=iTmVORK7NZ8G2hWxSf6R-c9BOGV7601VIiFhg7ZMZrc,1828
+livekit_plugins_flashtts-1.1.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+livekit_plugins_flashtts-1.1.0.dist-info/RECORD,,

livekit_plugins_flashtts-1.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.27.0
+Root-Is-Purelib: true
+Tag: py3-none-any