merge code repo for sensevoice and whisper

Dockerfile_sensevoice

@@ -0,0 +1,7 @@
+FROM corex:3.2.1
+
+WORKDIR /workspace
+COPY requirements_sensevoice.txt constraints_sensevoice.txt server_sensevoice.py launch_sensevoice.sh /workspace/
+RUN pip install -r requirements_sensevoice.txt -c constraints_sensevoice.txt
+
+ENTRYPOINT ["/bin/bash", "launch_sensevoice.sh"]

Dockerfile_whisper

@@ -0,0 +1,10 @@
+FROM zibo.harbor.iluvatar.com.cn:30000/saas/bi100-3.2.1-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.2
+
+RUN mkdir /workspace
+WORKDIR /workspace/
+
+COPY asr_server_whisper.py /workspace/asr_server_whisper.py
+COPY ./launch_service /workspace/launch_service
+COPY whisper-tiny /model
+
+ENTRYPOINT ["./launch_service"]

README.md

@@ -0,0 +1,6 @@
+# tiangai100-sensevoice-funasr
+# tiangai100-whisper
+
+【语音识别】
+
+funasr可正常照nvidia方式使用。

asr_server_whisper.py

@@ -0,0 +1,34 @@
+import torch
+from transformers import pipeline
+from flask import Flask, request, jsonify
+from threading import Lock
+
+asr_pipeline = pipeline(
+    task="automatic-speech-recognition",
+    model="/model",
+    torch_dtype=torch.float16,
+    device="cuda:0"
+)
+
+app = Flask(__name__)
+pipeline_lock = Lock()
+
+@app.route('/recognition', methods=['POST'])
+def predict():
+    audio_data = request.data
+    params = request.params
+    params_json = json.loads(params)
+    lang = params_json.get('language', 'en')
+    with pipeline_lock:
+        res = asr_pipeline(inputs=audio_data, generate_kwargs={"language": lang})
+    text = res['text'].strip()
+    return jsonify({'RecognitionStatus': 'Success', "DisplayText": text})
+
+@app.route('/ready', methods=['GET'])
+@app.route('/health', methods=['GET'])
+@app.route('/health_check', methods=['GET'])
+def health():
+    return jsonify({'status': 'ok'})
+
+if __name__ == '__main__':
+    app.run(host='0.0.0.0', port=80)

constraints_sensevoice.txt

@@ -0,0 +1 @@
+torch==2.1.0+corex.3.2.1

launch_sensevoice.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+python3 asr_server.py
+

launch_service

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+data
+cat /proc/cpuinfo
+ixsmi
+export
+date
+
+python3 asr.py

requirements_sensevoice.txt

@@ -0,0 +1,3 @@
+funasr
+fastapi
+uvicorn[standard]

server_sensevoice.py

@@ -0,0 +1,41 @@
+from fastapi import FastAPI, Request
+from contextlib import asynccontextmanager
+from funasr import AutoModel
+from funasr.utils.postprocess_utils import rich_transcription_postprocess
+import os
+import torch
+
+model_dir = os.getenv("MODEL_DIR", "/model/iic/SenseVoiceSmall")
+model = None
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    global model
+    if model is None:
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        model = AutoModel(model=model_dir, disable_update=True, device=device)
+    yield
+    pass
+
+app = FastAPI(lifespan=lifespan)
+
+@app.post("/recognition")
+async def asr(request: Request, language: str = "auto"):
+    audio_data = await request.body()
+    res = model.generate(input=audio_data, use_itn=True, ban_emo_unk=True)
+    text = rich_transcription_postprocess(res[0]["text"])
+    return {
+        "RecognitionStatus": "Success",
+        "DisplayText": text
+    }
+
+@app.get("/health")
+@app.get("/ready")
+def ready():
+    return {"status": "ok"}
+
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=80)
+

whisper-tiny/.mv

@@ -0,0 +1 @@
+Revision:master,CreatedAt:1736300537