如何将Flask请求FileStorage文件转换为numpy ndarray？

Question

我正在尝试将 FileStorage 对象转换为 numpy ndarray 以进行 ASR 转录。这是我的代码：

@app.route('/api/transcribe', methods=\['POST'\])
def transcribe():
   if 'audioFile' not in request.files:
       return 'No audio file provided', 400
   file = request.files\['audioFile'\]

接下来我必须将其转换为具有 16000 采样率和单通道的 ndarray。如何在不保存临时文件的情况下做到这一点？

我尝试使用声音文件、wav 或其他库，但没有任何用处。

Answer 1

要将 FileStorage 对象转换为 NumPy ndarray，您可以使用 Python 中的“io.BytesIO”模块。

def transcribe():
if 'audioFile' not in request.files:
    return 'No audio file provided', 400

file = request.files['audioFile']

# Read the file content as bytes
audio_data = file.read()

# Convert the bytes to a NumPy array using soundfile
with io.BytesIO(audio_data) as f:
    data, samplerate = sf.read(f)

# Check if the audio has multiple channels
if len(data.shape) > 1:
    # Convert to mono by averaging channels
    data = np.mean(data, axis=1)

# Resample the audio to 16kHz if necessary
if samplerate != 16000:
    # You may need to install the resampy package for this
    import resampy
    data = resampy.resample(data, samplerate, 16000)

# Ensure the data is in the correct dtype
data = data.astype(np.float32)

# Now you have your audio data in a NumPy array
print(data.shape)  # Shape of the array
print(samplerate)   # Sample rate of the audio

# Perform ASR transcription or other processing

return 'Transcription done', 200

请确保安装所需的软件包（soundfile 和 resampy）您可以使用 pip 安装它们：

pip install soundfile resampy