Low Volume

[joshs85]

Describe the feature you'd like
An environment variable to set the output volume level. The volume seems low on the output files.

[working-name]

I have no idea how any of this works, fiddled with the normalizer and whatever methods are documented without any results (audio sounds the same). You can however multiply chunk_audio (ndarray) by whatever gain ratio you want in generate_audio_stream() of tts_service.py. Not a good solution, since some intonations will raise volume above expected and it will clip.

                        customgain = 1.6 # good base gain for all voices
                        twopoint5 = ['af_sky'] # particularly low volume voice
                        twopoint4 = ['af_bella'] # almost as low
                        onepoint9 = ['af_sarah_af_sky'] # not as bad, but there was room past 1.6
                        if voice in twopoint5:
                            customgain = 2.5
                        if voice in twopoint4:
                            customgain = 2.4
                        if voice in onepoint9:
                            customgain = 1.9
...
chunk_bytes = AudioService.convert_audio(
                            chunk_audio * customgain, # this is where I added it
                            24000,
...

An update was released for kokoro with more voices but not sure if volume has also changed.

[joshs85]

I’m using af_heart from the update.  I downloaded it and dropped it in the folder but it seems low on the others also coming over from a piper voice.  I’ll give this a try later when I have some time.  On Feb 2, 2025, at 8:41 AM, Alan ***@***.***> wrote: I have no idea how any of this works, fiddled with the normalizer and whatever methods are documented without any results (audio sounds the same). You can however multiply chunk_audio (ndarray) by whatever gain ratio you want in generate_audio_stream() of tts_service.py. Not a good solution, since some intonations will raise volume above expected and it will clip. customgain = 1.6 # good base gain for all voices twopoint5 = ['af_sky'] # particularly low volume voice twopoint4 = ['af_bella'] # almost as low onepoint9 = ['af_sarah_af_sky'] # not as bad, but there was room past 1.6 ... chunk_bytes = AudioService.convert_audio( chunk_audio * customgain, # this is where I added it 24000, ... An update was released for kokoro with more voices but not sure if volume has also changed. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>