Potential other EndeavourOS Issue

[T1nyRaccoon]

Hi,

I might have found another EOS issue; at least the app does not seem to recognise any audio input even if I get very close to the mic and enunciate words.

I have tried manually changing audio input to my mic and their .monitor just in case but it doesnt seem to help
There is a possibility this is my fault as I have messed around a little trying to install coqui TTS models for example (as they dont show up in the download interface) and rerouting the audio output but havent touched audio input as far as I know.

Apologies for another issue and if it is user error.

Before the previous troubleshooting we had with TTS not working, it did register audio I think but was very bad (e.g i would say a sentence and three letters would come out) using the DeepSpeech Mozilla model

Apologies again for the inconvenience

[mkiol]

Apologies again for the inconvenience

Please don't. I am very grateful that you report potential errors.

I might have found another EOS issue; at least the app does not seem to recognise any audio input even if I get very close to the mic and enunciate words.

Please collect logs when starting "listening".

dsnote --verbose

DeepSpeech Mozilla model

While this has nothing to do with the error you reported, I would recommend using a more modern model such as WhisperCpp or Vosk. DeepSpeech is very old, inaccurate, and generally considered obsolete.

[T1nyRaccoon]

I have switched to Vosk, had the same issue. Honestly not sure why I didnt mention it.

This was me just repeating 'hello' into the mic

[D] 19:57:05.455244560.455 0x7fc9dd398d00 () - stt start listen
[D] 19:57:05.456218696.456 0x7fc9dd398d00 () - default mnt lang not found: "en"
[D] 19:57:05.456266117.456 0x7fc9dd398d00 () - new default mnt lang: "en"
[D] 19:57:05.456312395.456 0x7fc9dd398d00 () - choosing model for id: "en_vosk_large" "en"
[D] 19:57:05.456403649.456 0x7fc9dd398d00 () - restart stt engine config: "lang=en, lang_code=, model-files=[model-file=/home/t1nyraccoon/.cache/net.mkiol/dsnote/speech-models/en_vosk_large, scorer-file=, openvino-file=, ttt-model-file=], cache-dir=/home/t1nyraccoon/.cache/net.mkiol/dsnote, speech-mode=automatic, vad-mode=aggressiveness-3, speech-started=0, text-format=raw, cpu-threads=4, beam-search=5, audio-ctx-conf=dynamic, audio-ctx-conf-size=1500, options=, insert-stats=0, use-gpu=0, gpu-device=[id=-1, api=opencl, name=, platform-name=, flash-attn=0], sub-config=[min-segment-dur=4, min-line-length=0, max-line-length=0], translate=0, initial_prompt=0"
[D] 19:57:05.456839309.456 0x7fc9dd398d00 () - new stt engine not required, only restart
[D] 19:57:05.456899704.456 0x7fc9dd398d00 request_stop:280 - stt stop already requested
[D] 19:57:05.456959207.456 0x7fc9dd398d00 stop:299 - stt already stopped
[D] 19:57:05.457009513.457 0x7fc9dd398d00 start:254 - stt start
[D] 19:57:05.457091910.457 0x7fc9dd398d00 start:264 - stt start completed
[D] 19:57:05.457104905.457 0x7fc90d3ff680 process:313 - stt processing started
[D] 19:57:05.457201459.457 0x7fc90d3ff680 set_state:502 - stt state: idle => initializing
[D] 19:57:05.457140463.457 0x7fc9dd398d00 () - creating audio source
[D] 19:57:05.457252657.457 0x7fc90d3ff680 set_state:509 - speech detection status: no-speech => initializing (no-speech)
[D] 19:57:05.457391321.457 0x7fc90d3ff680 set_state:502 - stt state: initializing => idle
[D] 19:57:05.457456325.457 0x7fc90d3ff680 set_state:509 - speech detection status: initializing => no-speech (no-speech)
[D] 19:57:05.457300488.457 0x7fc9dd398d00 () - mic source created
[D] 19:57:05.755597442.755 0x7fc9dd398d00 () - using audio input: "alsa_output.pci-0000_03_00.1.hdmi-stereo-extra1.monitor" (preferred was "bluez_output.F4:9D:8A:CA:9F:D4.monitor" )
[D] 19:57:05.809291918.809 0x7fc9dd398d00 () - audio state: IdleState
[D] 19:57:05.809346532.809 0x7fc9dd398d00 set_speech_started:561 - speech started: false => true
[D] 19:57:05.809514101.809 0x7fc9dd398d00 () - service refresh status, new state: listening-auto
[D] 19:57:05.809558335.809 0x7fc9dd398d00 () - service state changed: idle => listening-auto
[D] 19:57:05.809872835.809 0x7fc9dd398d00 () - service refresh status, new state: listening-auto
[D] 19:57:05.809946645.809 0x7fc9dd398d00 () - service refresh status, new state: listening-auto
[D] 19:57:05.810064881.810 0x7fc9dd398d00 () - app current task: -1 => 1
[D] 19:57:05.810097653.810 0x7fc9dd398d00 () - app service state: idle => listening-auto
[W] 19:57:05.814031380.814 0x7fc9dd398d00 () - no available mnt langs
[W] 19:57:05.814073520.814 0x7fc9dd398d00 () - no available mnt out langs
[W] 19:57:05.814117153.814 0x7fc9dd398d00 () - no available tts models for in mnt
[W] 19:57:05.814160897.814 0x7fc9dd398d00 () - no available tts models for out mnt
[D] 19:57:05.816369635.816 0x7fc9dd398d00 () - [dbus app] State called
[D] 19:57:06.018890329.18 0x7fc9dd398d00 () - audio state: ActiveState
[D] 19:57:07.279102519.279 0x7fc90d3ff680 process_buff:226 - process samples buf: mode=automatic, in-buf size=24000, speech-buf size=0, sof=true, eof=false
[D] 19:57:07.314959482.314 0x7fc90d3ff680 process_buff:290 - vad: no speech
[D] 19:57:08.879200707.879 0x7fc90d3ff680 process_buff:226 - process samples buf: mode=automatic, in-buf size=24000, speech-buf size=0, sof=false, eof=false
[D] 19:57:08.889096996.889 0x7fc90d3ff680 process_buff:290 - vad: no speech
[D] 19:57:10.279058646.279 0x7fc90d3ff680 process_buff:226 - process samples buf: mode=automatic, in-buf size=24000, speech-buf size=0, sof=false, eof=false
[D] 19:57:10.288905682.288 0x7fc90d3ff680 process_buff:290 - vad: no speech
[D] 19:57:11.878879749.878 0x7fc90d3ff680 process_buff:226 - process samples buf: mode=automatic, in-buf size=24000, speech-buf size=0, sof=false, eof=false
[D] 19:57:11.888768574.888 0x7fc90d3ff680 process_buff:290 - vad: no speech

As an aside, I was unable to find larger neural TTS models such as Coqui on the inbuilt installer and struggled for a while before eventually giving up on installing and adding a decent neural model. Is this just due to me being on a Wayland session? or have those larger models simply not been added yet?

Thanks for your help.

[mkiol]

[D] 19:57:05.755597442.755 0x7fc9dd398d00 () - using audio input: "alsa_output.pci-0000_03_00.1.hdmi-stereo-extra1.monitor" (preferred was "bluez_output.F4:9D:8A:CA:9F:D4.monitor" )

It appears that the chosen audio source is "alsa_output.pci-0000_03_00.1.hdmi-stereo-extra1.monitor", which is not a microphone.

For some reason, the Speech Note does not select the microphone. One possible explanation is that no mic is connected. Is it connected? You can manually set the audio source in the settings (Speech To Text > Audio input device).

After launching the application, look for the entry "source_info_pa_callback:206 - pa source:" in the log. All detected audio input devices should be listed there.

[T1nyRaccoon]

That's really odd, it was set to the monitor of my headphones, although I am fairly confident I did not set it as that. (Then again I may have realistically misclicked lol)

Sorry for that.

If you dont mind I will ask this as it states on the github page that these models should be available, yet I couldnt find them

I was unable to find larger neural TTS models such as Coqui on the inbuilt installer and struggled for a while before eventually giving up on installing and adding a decent neural model. Is this just due to me being on a Wayland session? or have those larger models simply not been added yet?

[mkiol]

it was set to the monitor of my headphones

You most likely do not want to select "monitor". Monitor is captured audio playback from your speakers. It makes sense to set the audio input to the monitor device only if you want to apply STT on audio from your computer - for example, you can transcribe audio from videos played on YouTube, etc. If you want to transcribe your own voice, you should select a microphone.

If you dont mind I will ask this as it states on the github page that these models should be available, yet I couldnt find them

Everything is as expected. Arch package does not include the Python libraries required to support additional engines, so some models are hidden. You must install the Python dependencies separately somewhere on your system. The most convenient way is to create a separate Python venv. After creation, you must set the venv directory in Settings > Advanced > Location of Python libraries. Below is my configuration. I created venv in /home/mkiol/dev/speechnote-venv.

A couple of examples of python pip packages to install:

• coqui-tts (Coqui TTS engine - I recommend play with Coqui XTTS voice cloning model)
• faster-whisper (FasterWhisper STT engine)
• parler-tts (Parler TTS engine)
• f5-tts (F5 TTS engine - also good for voice cloning)
• kokoro (Kokoro TTS engine)

As an alternative to installing the AUR package, you can simply install the Flatpak package. The Flatpak package contains all dependencies, so you don't need to install anything else on your system.