一、创建FunASR的API服务便于联网调用

计划与准备

使用 Flask 创建的 Python API，它接受音频流（作为上传文件），利用 funasr 库和 SenseVoiceSmall 模型进行转录，并返回识别出的文字。该 API 将持续运行直到你停止服务器。

需要安装了 funasr 及其依赖（如 PyTorch 和适当的 CUDA 环境如果使用 GPU），并且 SenseVoiceSmall 模型可用或可以通过 funasr 下载。

实现步骤：

1. 设置一个 Flask Web 服务器。
2. funasr 模型将在服务器启动时加载一次，以节省后续请求的时间。
3. 创建一个 API 端点（例如 /transcribe），接受 POST 请求。
4. 该端点期望请求中包含音频文件。
5. 收到的音频文件将被暂时保存，然后由 funasr 模型处理。
6. 转录出的文字将以 JSON 响应返回。

FunASR 语音识别 API的代码

import os
import tempfile
from flask import Flask, request, jsonify
from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess

# --- Configuration ---
# Specify the model directory or model ID that AutoModel can download.
# If "SenseVoiceSmall" is a local directory, ensure it's accessible.
# If it's a model ID, AutoModel will attempt to download it.
MODEL_DIR = "SenseVoiceSmall"
# Specify the device: "cuda:0" for GPU (if available), or "cpu".
# Ensure you have the necessary CUDA drivers and PyTorch version if using GPU.
DEVICE = "cuda:0" # or "cpu"

# --- Initialize Flask App ---
app = Flask(__name__)

# --- Load FunASR Model ---
# This model is loaded once when the Flask app starts.
print(f"Loading FunASR model: {MODEL_DIR} on device: {DEVICE}...")
try:
    model = AutoModel(
        model=MODEL_DIR,
        vad_model="fsmn-vad",
        vad_kwargs={"max_single_segment_time": 30000},
        device=DEVICE,
    )
    print("FunASR model loaded successfully.")
except Exception as e:
    print(f"Error loading FunASR model: {e}")
    # If the model fails to load, you might want to exit or handle this gracefully.
    # For this example, we'll let Flask start but the /transcribe endpoint will fail.
    model = None

@app.route('/transcribe', methods=['POST'])
def transcribe_audio():
    """
    API endpoint to transcribe an uploaded audio file.
    Expects a POST request with a file part named 'audio_file'.
    """
    if model is None:
        return jsonify({"error": "Model not loaded. Check server logs."}), 500

    if 'audio_file' not in request.files:
        return jsonify({"error": "No audio file part in the request"}), 400

    audio_file = request.files['audio_file']

    if audio_file.filename == '':
        return jsonify({"error": "No selected audio file"}), 400

    if audio_file:
        try:
            # Create a temporary file to save the uploaded audio
            # tempfile.NamedTemporaryFile creates a file that is deleted when closed.
            # We need to pass the filename string to funasr, so we'll manage deletion manually.
            temp_audio_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav") # You can adjust suffix if needed
            audio_file.save(temp_audio_file.name)
            temp_audio_path = temp_audio_file.name
            temp_audio_file.close() # Close the file handle

            print(f"Temporary audio file saved at: {temp_audio_path}")

            # Perform transcription
            # The cache parameter can be an empty dict or a more sophisticated caching object
            # if you expect to process the same audio multiple times.
            res = model.generate(
                input=temp_audio_path,
                cache={},
                language="auto",  # Automatically detect language
                use_itn=True,     # Apply Inverse Text Normalization
                batch_size_s=60,  # Batch size in seconds
                merge_vad=True,   # Merge VAD segments
                merge_length_s=15 # Merge segments up to this length in seconds
            )

            if not res or "text" not in res[0]:
                return jsonify({"error": "Transcription failed or returned unexpected result"}), 500

            # Post-process the transcription for rich formatting (if applicable)
            transcribed_text = rich_transcription_postprocess(res[0]["text"])

            return jsonify({"transcription": transcribed_text})

        except Exception as e:
            print(f"Error during transcription: {e}")
            return jsonify({"error": str(e)}), 500
        finally:
            # Clean up the temporary file
            if 'temp_audio_path' in locals() and os.path.exists(temp_audio_path):
                os.remove(temp_audio_path)
                print(f"Temporary audio file {temp_audio_path} deleted.")
    else:
        return jsonify({"error": "Audio file processing failed"}), 400

if __name__ == '__main__':
    # Run the Flask app
    # For development, host='0.0.0.0' makes it accessible on your network.
    # For production, use a proper WSGI server like Gunicorn or uWSGI.
    app.run(host='0.0.0.0', port=5000, debug=True)

使用 funasr 的语音转文本提供了一个功能性的 API。请根据你的环境调整模型路径和设备设置。API 将持续活跃，直到你手动停止 Python 脚本。