|
| 1 | +#!/usr/bin/env python3 |
| 2 | +""" |
| 3 | +gRPC backend for LocalAI wrapping FunASR (SenseVoice / Paraformer). |
| 4 | +""" |
| 5 | +from concurrent import futures |
| 6 | +import time |
| 7 | +import argparse |
| 8 | +import signal |
| 9 | +import sys |
| 10 | +import os |
| 11 | +import backend_pb2 |
| 12 | +import backend_pb2_grpc |
| 13 | +import torch |
| 14 | + |
| 15 | +import grpc |
| 16 | +sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..', 'common')) |
| 17 | +sys.path.insert(0, os.path.join(os.path.dirname(__file__), 'common')) |
| 18 | +from grpc_auth import get_auth_interceptors |
| 19 | + |
| 20 | + |
| 21 | +_ONE_DAY_IN_SECONDS = 60 * 60 * 24 |
| 22 | +MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1')) |
| 23 | + |
| 24 | + |
| 25 | +class BackendServicer(backend_pb2_grpc.BackendServicer): |
| 26 | + def Health(self, request, context): |
| 27 | + return backend_pb2.Reply(message=bytes("OK", 'utf-8')) |
| 28 | + |
| 29 | + def LoadModel(self, request, context): |
| 30 | + from funasr import AutoModel |
| 31 | + |
| 32 | + device = "cpu" |
| 33 | + if request.CUDA and torch.cuda.is_available(): |
| 34 | + device = "cuda" |
| 35 | + mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available() |
| 36 | + if mps_available: |
| 37 | + device = "mps" |
| 38 | + |
| 39 | + model_id = request.Model or "iic/SenseVoiceSmall" |
| 40 | + |
| 41 | + try: |
| 42 | + print(f"Loading FunASR model: {model_id} on {device}", file=sys.stderr) |
| 43 | + self.model = AutoModel( |
| 44 | + model=model_id, |
| 45 | + vad_model="fsmn-vad", |
| 46 | + device=device, |
| 47 | + disable_update=True, |
| 48 | + ) |
| 49 | + print("FunASR model loaded successfully", file=sys.stderr) |
| 50 | + except Exception as err: |
| 51 | + print(f"[ERROR] LoadModel failed: {err}", file=sys.stderr) |
| 52 | + import traceback |
| 53 | + traceback.print_exc(file=sys.stderr) |
| 54 | + return backend_pb2.Result(success=False, message=str(err)) |
| 55 | + |
| 56 | + return backend_pb2.Result(message="Model loaded successfully", success=True) |
| 57 | + |
| 58 | + def AudioTranscription(self, request, context): |
| 59 | + result_segments = [] |
| 60 | + text = "" |
| 61 | + try: |
| 62 | + audio_path = request.dst |
| 63 | + if not audio_path or not os.path.exists(audio_path): |
| 64 | + print(f"Error: Audio file not found: {audio_path}", file=sys.stderr) |
| 65 | + return backend_pb2.TranscriptResult(segments=[], text="") |
| 66 | + |
| 67 | + language = None |
| 68 | + if request.language and request.language.strip(): |
| 69 | + language = request.language.strip() |
| 70 | + |
| 71 | + kwargs = {} |
| 72 | + if language: |
| 73 | + kwargs["language"] = language |
| 74 | + |
| 75 | + results = self.model.generate(input=audio_path, **kwargs) |
| 76 | + |
| 77 | + if not results: |
| 78 | + return backend_pb2.TranscriptResult(segments=[], text="") |
| 79 | + |
| 80 | + for idx, r in enumerate(results): |
| 81 | + seg_text = r.get("text", "") if isinstance(r, dict) else str(r) |
| 82 | + text += seg_text |
| 83 | + result_segments.append(backend_pb2.TranscriptSegment( |
| 84 | + id=idx, |
| 85 | + start=0, |
| 86 | + end=0, |
| 87 | + text=seg_text, |
| 88 | + )) |
| 89 | + |
| 90 | + except Exception as err: |
| 91 | + print(f"Error in AudioTranscription: {err}", file=sys.stderr) |
| 92 | + import traceback |
| 93 | + traceback.print_exc(file=sys.stderr) |
| 94 | + return backend_pb2.TranscriptResult(segments=[], text="") |
| 95 | + |
| 96 | + return backend_pb2.TranscriptResult(segments=result_segments, text=text) |
| 97 | + |
| 98 | + |
| 99 | +def serve(address): |
| 100 | + server = grpc.server( |
| 101 | + futures.ThreadPoolExecutor(max_workers=MAX_WORKERS), |
| 102 | + options=[ |
| 103 | + ('grpc.max_message_length', 50 * 1024 * 1024), |
| 104 | + ('grpc.max_send_message_length', 50 * 1024 * 1024), |
| 105 | + ('grpc.max_receive_message_length', 50 * 1024 * 1024), |
| 106 | + ], |
| 107 | + interceptors=get_auth_interceptors(), |
| 108 | + ) |
| 109 | + backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server) |
| 110 | + server.add_insecure_port(address) |
| 111 | + server.start() |
| 112 | + print("Server started. Listening on: " + address, file=sys.stderr) |
| 113 | + |
| 114 | + def signal_handler(sig, frame): |
| 115 | + print("Received termination signal. Shutting down...") |
| 116 | + server.stop(0) |
| 117 | + sys.exit(0) |
| 118 | + |
| 119 | + signal.signal(signal.SIGINT, signal_handler) |
| 120 | + signal.signal(signal.SIGTERM, signal_handler) |
| 121 | + |
| 122 | + try: |
| 123 | + while True: |
| 124 | + time.sleep(_ONE_DAY_IN_SECONDS) |
| 125 | + except KeyboardInterrupt: |
| 126 | + server.stop(0) |
| 127 | + |
| 128 | + |
| 129 | +if __name__ == "__main__": |
| 130 | + parser = argparse.ArgumentParser(description="Run the gRPC server.") |
| 131 | + parser.add_argument("--addr", default="localhost:50051", help="The address to bind the server to.") |
| 132 | + args = parser.parse_args() |
| 133 | + serve(args.addr) |
0 commit comments