Cognitive-services-speech-sdk: Speech to text conversion seems to handle max 15~16 seconds of audio

[viveknirkhe]

Please provide us with the following information:

This issue is for a: (mark with an x)

- [x] bug report -> please search issues before submitting
- [ ] feature request
- [ ] documentation issue or request
- [ ] regression (a behavior that used to work and stopped in a new release)

Minimal steps to reproduce

>
Select any audio wav file longer than 16 seconds. And tried converted files using both option 3 and option 6, but audio of upto 15~16 seconds is recognized and converted into text. The rest of audio is ignored. There is no error message.

Any log messages given by the failure

>

Expected/desired behavior

>
Entire audio file should be converted.

OS and Version?

Windows 7, 8 or 10. Linux (which distribution). Other.
Windows 10 pro

Versions

>

Mention any other details that might be useful

The REST API does mention that it handles max of 15 seconds but longer audio is handled by the SDK. I am using SDK.

---

Thanks! We'll be in touch soon.

[zhouwangzw]

This is the current limitation. We are working on it and will remove this limitation very soon.

[viveknirkhe]

Is this the limitation in the REST API or the library?

[zhouwangzw]

REST API has similar limitation. In future, we are going to remove this limitation in SDK very soon.

[bluemonk482]

Any updates on this. Otherwise you should change your documentation here.

[zhouwangzw]

We are working on it. The next coming release in the next 1-2 weeks will support long-running recognition. Sorry for the inconvenience.

[mahilleb-msft]

@viveknirkhe, @bluemonk482 - our new 0.5.0 release that should fix this is out now.
Please check out the updated NuGet package https://www.nuget.org/packages/Microsoft.CognitiveServices.Speech and the release notes.

Thanks, Mark

[mahilleb-msft]

Closing. Please open another issue if things do not work as expected.

Thanks, Mark

[lwluc]

Using JS-Browser Quickstart and reading a .wav file with SpeechSDK.AudioConfig.fromWavFileInput(file). The transcriptions stops after 15 seconds. I am not getting an error, it just stops. I am using the 7-day test version and speechRecognitionLanguage = "de-DE" does this have any impact?
Am I doing something wrong?

[wolfma61]

Hey @lwluc -
as Mark said, please open a new issue for new and unrelated questions in the future.

The sample is using RecognizeOnce ... that limits recognition to 10-15 seconds
see https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/speechrecognizer?view=azure-node-latest
"recognizeonceasync: recognizeOnceAsync:
Starts speech recognition, and stops after the first utterance is recognized. "

for long running audio you will need to utilize Start/StopRecognize and subscribe to recognition events.
https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/speechrecognizer?view=azure-node-latest#startcontinuousrecognitionasync

thx
Wolfgang

[kiranmahto]

sdk for longer audio is in c# is it for python also

[chlandsi]

sdk for longer audio is in c# is it for python also

Here is some example code on how to use continuous recognition on audio files of arbitrary length:

```python
def speech_recognize_continuous_from_file():
"""performs continuous speech recognition with input from an audio file"""
speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)
audio_config = speechsdk.audio.AudioConfig(filename=audiofilename)

speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)

done = False

def stop_cb(evt):
    """callback that stops continuous recognition upon receiving an event `evt`"""
    print('CLOSING on {}'.format(evt))
    speech_recognizer.stop_continuous_recognition()
    nonlocal done
    done = True

# Connect callbacks to the events fired by the speech recognizer
speech_recognizer.recognizing.connect(lambda evt: print('RECOGNIZING: {}'.format(evt)))
speech_recognizer.recognized.connect(lambda evt: print('RECOGNIZED: {}'.format(evt)))
speech_recognizer.session_started.connect(lambda evt: print('SESSION STARTED: {}'.format(evt)))
speech_recognizer.session_stopped.connect(lambda evt: print('SESSION STOPPED {}'.format(evt)))
speech_recognizer.canceled.connect(lambda evt: print('CANCELED {}'.format(evt)))
# stop continuous recognition on either session stopped or canceled events
speech_recognizer.session_stopped.connect(stop_cb)
speech_recognizer.canceled.connect(stop_cb)

# Start continuous speech recognition
speech_recognizer.start_continuous_recognition()
while not done:
    time.sleep(.5)

```

[LeonRomaniuk]

Yes, our latest release of the Speech SDKhttps://azure.microsoft.com/en-us/services/cognitive-services/directory/speech/ now supports Python. Thank you for your interest.

From: kiranmahto [e-mail]
Sent: Thursday, January 10, 2019 9:20 PM
To: Azure-Samples/cognitive-services-speech-sdk [e-mail]
Cc: Subscribed [e-mail]
Subject: Re: [Azure-Samples/cognitive-services-speech-sdk] Speech to text conversion seems to handle max 15~16 seconds of audio (#13)

sdk for longer audio is in c# is it for python also

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHubhttps://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FAzure-Samples%2Fcognitive-services-speech-sdk%2Fissues%2F13%23issuecomment-453381908&data=02%7C01%7CLeonro%40microsoft.com%7Cfcd4da394d9e402620dc08d677847601%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636827808151213199&sdata=Ps899ak0M3RJz%2BCuPKgBfJI9Kn4%2FCry901yzUpvhEzY%3D&reserved=0, or mute the threadhttps://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FArVXa4U_MmM22dA0SwNGy0FXCJCkVCq_ks5vCB8LgaJpZM4UqVEN&data=02%7C01%7CLeonro%40microsoft.com%7Cfcd4da394d9e402620dc08d677847601%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636827808151223203&sdata=J2tMs02hCbY3D85wt4toAXkoOhMhZX0kMNfGow975kg%3D&reserved=0.

[LeonRomaniuk]

Leon Romaniuk would like to recall the message, "[Azure-Samples/cognitive-services-speech-sdk] Speech to text conversion seems to handle max 15~16 seconds of audio (#13)".

[kiranmahto]

Thanks @chlandsi , @LeonRomaniuk

[varuntahin2210]

Is there a way to tag speakers in the above code provided by @chlandsi ? Please help

[wolfma61]

you mean identify speakers?
this is currently not supported.

[JamesEarle]

for long running audio you will need to utilize Start/StopRecognize and subscribe to recognition events.
https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/speechrecognizer?view=azure-node-latest#startcontinuousrecognitionasync

@wolfma61 does the SDK support continuous recognition from live audio, similar to RecognizeOnce? Or only from audio files?

[wolfma61]

works for all audio input, so files, streams and microphone as long as you use continuous recognition (not RecognizeOnce)
thx
Wolfgang

[JamesEarle]

@wolfma61 are there any code samples showing this here?

[wolfma61]

The sample directory contains sample that do call StartContinuousRecognitionAsync
If they do it from a file, just don't use the file input, default is microphone ...