#CognitiveServices – How to create audio files for Custom Speech Service (#CRIS)

Hello!

A few days ago I was asked about an easy way to create audio files to be used as datasets in Custom Speech Service (CRIS). As I mentioned in a previous post, the audio files must have special features, so it is important to create them correctly.

Note: the files are WAVs files, mono and another pair of details makes it not easy to create them in a single step.

Although there are several ways to create these files, this is the one I use and it works.

• To record the audio I use an app that comes by default in Windows: Voice Recorder

• I guess I don’t need to explain how the app works. Just press the microphone button. Nor do we expect many options in the Settings section.

• Once we have recorded a session, we can access the list of recordings. If we see the record path of the file we will see that it is recorded with the name “Recording.m4a”

• Now is the time to find a way to convert M4A files to WAV. In this case I use VLC (link). The software is well known, so I will not write a lot about it. In VLC Select the option “Media // Convert / Save …”

• Select a file and press the option “Convert”

• In this step we must create a profile with the information needed to create compatible CRIS compatible files.
• I created a profile called “WAV Cris 02” with the following configurations
• Encapsulation: WAV

• Audio codec with the values required by CRIS

• Now we can use this profile to convert our M4A file to WAV

• Ready! We have a WAV file which is compatible with CRIS requirements and we can use the file for our data models.

Happy coding ! 😀

Saludos @ Burlington

El Bruno

References

• El Bruno
  • Sample Console App to perform Audio analysis using Custom Speech Service
  • Tutorial to create and publish a complete model in Custom Speech Service (#CRIS)
• Cognitive Services, Custom Speech Service
• VLC, VideoLan