SOTA Open Source TTS
Fix tools.api_server fails to run problem (#1011)
cronrpc352d076db7e0
diff --git a/tools/server/api_utils.py b/tools/server/api_utils.pyindex dfe9900..ea99896 100644--- a/tools/server/api_utils.py+++ b/tools/server/api_utils.py@@ -13,7 +13,7 @@ from tools.server.inference import inference_wrapper as inferencedef parse_args():parser = ArgumentParser()- parser.add_argument("--mode", type=str, choices=["agent", "tts"], default="tts")+ parser.add_argument("--mode", type=str, choices=["tts"], default="tts")parser.add_argument("--load-asr-model", action="store_true")parser.add_argument("--llama-checkpoint-path",diff --git a/tools/server/model_manager.py b/tools/server/model_manager.pyindex 77540a2..25ac3ad 100644--- a/tools/server/model_manager.py+++ b/tools/server/model_manager.py@@ -4,10 +4,7 @@ from loguru import loggerfrom fish_speech.inference_engine import TTSInferenceEnginefrom fish_speech.models.dac.inference import load_model as load_decoder_model-from fish_speech.models.text2semantic.inference import (- launch_thread_safe_queue,- launch_thread_safe_queue_agent,-)+from fish_speech.models.text2semantic.inference import launch_thread_safe_queuefrom fish_speech.utils.schema import ServeTTSRequestfrom tools.server.inference import inference_wrapper as inference@@ -84,15 +81,6 @@ class ModelManager:precision=precision,compile=compile,)- elif mode == "agent":- self.llama_queue, self.tokenizer, self.config = (- launch_thread_safe_queue_agent(- checkpoint_path=checkpoint_path,- device=device,- precision=precision,- compile=compile,- )- )else:raise ValueError(f"Invalid mode: {mode}")diff --git a/tools/server/views.py b/tools/server/views.pyindex 905f6ad..210a5aa 100644--- a/tools/server/views.py+++ b/tools/server/views.py@@ -22,14 +22,12 @@ from typing_extensions import Annotatedfrom fish_speech.utils.schema import (ServeASRRequest,ServeASRResponse,- ServeChatRequest,ServeTTSRequest,ServeVQGANDecodeRequest,ServeVQGANDecodeResponse,ServeVQGANEncodeRequest,ServeVQGANEncodeResponse,)-from tools.server.agent import get_response_generatorfrom tools.server.api_utils import (buffer_to_async_generator,get_content_type,@@ -130,7 +128,7 @@ async def tts(req: Annotated[ServeTTSRequest, Body(exclusive=True)]):app_state = request.app.statemodel_manager: ModelManager = app_state.model_managerengine = model_manager.tts_inference_engine- sample_rate = engine.decoder_model.spec_transform.sample_rate+ sample_rate = engine.decoder_model.sample_rate# Check if the text is too longif app_state.max_text_length > 0 and len(req.text) > app_state.max_text_length:@@ -172,42 +170,3 @@ async def tts(req: Annotated[ServeTTSRequest, Body(exclusive=True)]):},content_type=get_content_type(req.format),)--[email protected]("/v1/chat")-async def chat(req: Annotated[ServeChatRequest, Body(exclusive=True)]):- # Check that the number of samples requested is correct- if req.num_samples < 1 or req.num_samples > MAX_NUM_SAMPLES:- raise HTTPException(- HTTPStatus.BAD_REQUEST,- content=f"Number of samples must be between 1 and {MAX_NUM_SAMPLES}",- )-- # Get the type of content provided- content_type = request.headers.get("Content-Type", "application/json")- json_mode = "application/json" in content_type-- # Get the models from the app- model_manager: ModelManager = request.app.state.model_manager- llama_queue = model_manager.llama_queue- tokenizer = model_manager.tokenizer- config = model_manager.config-- device = request.app.state.device-- # Get the response generators- response_generator = get_response_generator(- llama_queue, tokenizer, config, req, device, json_mode- )-- # Return the response in the correct format- if req.streaming is False:- result = response_generator()- if json_mode:- return JSONResponse(result.model_dump())- else:- return ormsgpack.packb(result, option=ormsgpack.OPT_SERIALIZE_PYDANTIC)-- return StreamResponse(- iterable=response_generator(), content_type="text/event-stream"- )