how to recognize a wav file?

[LindgeW]

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

[jm12138]

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.getframerate()
nframes = wav_file.getnframes()

# Read PCM data
pcm_data = wav_file.readframes(nframes)
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 frames, 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"))

[LindgeW]

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.getframerate()
nframes = wav_file.getnframes()

Read PCM data

pcm_data = wav_file.readframes(nframes)
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 frames, 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"))

Thanks for your reply! I have run it but an error was occurred as follows:

File "D:\Drivers\MyInstall\anaconda3\Lib\site-packages\msc\__init__.py", line 56, in asr epStatus, recogStatus = self.QISRAudioWrite( ^^^^^^^^^^^^^^^^^^^^ File "D:\Drivers\MyInstall\anaconda3\Lib\site-packages\msc\qisr.py", line 122, in QISRAudioWrite MSPAssert(errorCode, "QISRAudioWrite failed") File "D:\Drivers\MyInstall\anaconda3\Lib\site-packages\msc\msp.py", line 1036, in MSPAssert assert errorCode == MSPStatus.MSP_SUCCESS, "%s, error code: %d" % ( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ AssertionError: QISRAudioWrite failed, error code: 10106

[jm12138]

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.getframerate()
nframes = wav_file.getnframes()

Read PCM data

pcm_data = wav_file.readframes(nframes)
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 frames, 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"))

Thanks for your reply! I have run it but an error was occurred as follows:

File "D:\Drivers\MyInstall\anaconda3\Lib\site-packages\msc\__init__.py", line 56, in asr epStatus, recogStatus = self.QISRAudioWrite( ^^^^^^^^^^^^^^^^^^^^ File "D:\Drivers\MyInstall\anaconda3\Lib\site-packages\msc\qisr.py", line 122, in QISRAudioWrite MSPAssert(errorCode, "QISRAudioWrite failed") File "D:\Drivers\MyInstall\anaconda3\Lib\site-packages\msc\msp.py", line 1036, in MSPAssert assert errorCode == MSPStatus.MSP_SUCCESS, "%s, error code: %d" % ( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ AssertionError: QISRAudioWrite failed, error code: 10106

Please refer to the official error code guide

错误码 错误码解释
10106 无效的参数

报错原因
可能的原因如下：
· 参数名与文档不符
· 参数值不在规定范围内
· (WebAPI)编码引起的参数问题
· (WebAPI)对body的参数没有进行urlencode处理

解决方案
1.按照文档说明，输入正确的参数名称
2. 离线能力报错10106，比如resPath: null，可能是离线资源没有放，资源不对，或者路径没放对。实质是资源没有正确加载到导致。

论坛参考 http://bbs.xfyun.cn/forum.php?mod=viewthread&tid=17520&extra=
http://bbs.xfyun.cn/forum.php?mod=viewthread&tid=10134&highlight=10106
http://bbs.xfyun.cn/forum.php?mod=viewthread&tid=891&highlight=10106

[LindgeW]

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.getframerate()
nframes = wav_file.getnframes()

Read PCM data

pcm_data = wav_file.readframes(nframes)
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 frames, 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"))

Thanks for your reply. I notice that the parameter type of stream for client.asr is actually pyaudio.PyAudio.Stream, which does not match pcm_stream = io.BytesIO(resampled_audio.tobytes()). I suspect that this may lead to above bugs.

iFLYTEK-MSC-Python-SDK/msc/__init__.py

Line 39 in 1f37bbd

self, params: bytes, stream: Stream, chunk_size: int = 2048