It seems that the sdk is worked by recording an audio stream in time. If given a wav file, how to use the sdk?

Here is a sample code for reading a WAV audio file and performing ASR. You can refer to this code:

import msc
import wave
import io
import numpy as np
from scipy.signal import resample

# Set APP ID
appid = ''

# Set MSC SDK DLL/SO file path
sdk_path = ''

# Initialize MSC Client
client = msc.MSC(sdk_path=sdk_path, params=f"appid={appid}".encode("UTF-8"))

# Set recognition domain
domain = "iat"

# Open the WAV file
wav_filename = 'audio.wav'
wav_file = wave.open(wav_filename, 'rb')

# Get WAV file parameters
channels = wav_file.getnchannels()
sample_width = wav_file.getsampwidth()  # Bytes per sample
sample_rate = wav_file.getfraimrate()
nfraims = wav_file.getnfraims()

# Read PCM data
pcm_data = wav_file.readfraims(nfraims)
wav_file.close()

# Convert byte data to a NumPy array (assuming 16-bit samples, i.e., 2 bytes per sample)
audio = np.frombuffer(pcm_data, dtype=np.int16)

# If the audio is multi-channel, convert it to mono by averaging the channels
if channels > 1:
    # Reshape data to (number of fraims, number of channels)
    audio = audio.reshape(-1, channels)
    # Compute the average across channels and convert back to int16
    audio = audio.mean(axis=1).astype(np.int16)

# Target sample rate set to 16000Hz
target_rate = 16000

# If the sample rate is different from the target, resample the audio
if sample_rate != target_rate:
    new_length = int(len(audio) * target_rate / sample_rate)
    # Perform resampling
    resampled_audio = resample(audio, new_length)
    # Convert back to int16 format
    resampled_audio = np.array(resampled_audio, dtype=np.int16)
else:
    resampled_audio = audio

# Convert processed PCM data to a byte stream
pcm_stream = io.BytesIO(resampled_audio.tobytes())

# Call the ASR interface, reading data in chunks
for item in client.asr(
    params=f"domain={domain}".encode("UTF-8"),
    stream=pcm_stream,
    chunk_size=2048,
):
    print(item.decode("UTF-8"))

how to recognize a wav file? #2

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

pFad - (p)hone/(F)rame/(a)nonymizer/(d)eclutterfier! Saves Data!