Menu

PocketSphinx with Bluetooth Microphone

Help
eLar19
2020-03-05
2020-03-07
  • eLar19

    eLar19 - 2020-03-05

    Hi guys, I've build a pocketsphinx project with python, it uses LiveSpeech for continuous recognition. It works very well with the microphone of my laptop, also with my earphones connected to jack, but the finally product require to use a Bluetooth Headset. I tried a lot of Bluetooth Headset, but every of them has the same problem, the recognition is very very very bad. If with cable (or microphone of my 10 years old laptop) the recognition works well with 80% word recognized, with bluetooth the results are 10% word recognized.

    Any idea?

     
    • Nickolay V. Shmyrev

      You can share audio samples but most likely your bluetooth records have 8khz bandwidth and require 8khz model

      You can also get much better accuracy with vosk-api instead of livespeech.

       
      • eLar19

        eLar19 - 2020-03-06

        I tried to install vosk-api but without luck, It doen's work. In any case, I'm running the python script on linux, and I think that bluetooth managing on linux is not so good, so I tried to port it on Windows, but when I start the script with LiveSpeech, on Windows it recognize only 1 word per phrase, with bad accuracy. It's strange because the script is the same, I've installed latest pocketsphinx on python 3.7, same language module and dictionary. What's wrong?

         
        • Nickolay V. Shmyrev

          On Windows install vosk-api with

          pip3 install https://github.com/alphacep/vosk-api/releases/download/0.3.3/vosk-0.3.3-cp37-cp37m-win_amd64.whl

          To get help on the accuracy share the audio file recorded from your microphone to reproudce accuracy issues.

           
          • eLar19

            eLar19 - 2020-03-06

            I attached an audio file recordered with Audacity. There is a way to install vosk-api in linux? And how it works? Vosk uses kaldi I guess, so pocketsphinx will not be used anymore with this api?

             
            • Nickolay V. Shmyrev

              You audio is 8khz, you can try with 8khz model to get a good recognition accuracy.

               
              • eLar19

                eLar19 - 2020-03-07

                I tried to use the 8khz model, on LiveSpeech I setted "sampling_rate = 8000" insted of 16000, but the results are the same, not good.

                 
          • eLar19

            eLar19 - 2020-03-06

            I tried with Linux with your github repository, I launched the example with your wav and it worked fine this time. But how can I use my own model, with my dict and my lm?

             
  • Yanay Lehavi

    Yanay Lehavi - 2020-03-06

    With the latest AirPods we experience nearly identical accuracy between wired and Bluetooth.

     

Log in to post a comment.

Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.