Pipeline seems to randomly hang after loading the wav2txt model

[ethereal-engineer]

I'd like to get more debugging information on this but the last line (on a full pipeline running locally) is INFO:

DEBUG:rhasspy3.core:Loading config from /home/doc/rhasspy3/rhasspy3/configuration.yaml
DEBUG:rhasspy3.core:Loading config from /home/doc/rhasspy3/config/configuration.yaml
DEBUG:rhasspy3.program:mic_adapter_raw.py ['--samples-per-chunk', '1024', '--rate', '16000', '--width', '2', '--channels', '1', 'arecord -q -D "default" -r 16000 -c 1 -f S16_LE -t raw -']
DEBUG:rhasspy3.program:asr_adapter_wav2text.py ['script/wav2text --language en "/home/doc/rhasspy3/config/data/asr/faster-whisper/tiny-int8" "{wav_file}"']
DEBUG:rhasspy3.program:.venv/bin/python3 ['bin/porcupine_stream.py', '--model', 'grasshopper_linux.ppn']
DEBUG:rhasspy3.wake:detect: processing audio
DEBUG:rhasspy3.wake:detect: Detection(name='grasshopper_linux', timestamp=17743906487598)
DEBUG:rhasspy3.program:vad_adapter_raw.py ['--rate', '16000', '--width', '2', '--channels', '1', '--samples-per-chunk', '512', 'script/speech_prob "share/silero_vad.onnx"']
DEBUG:rhasspy3.vad:segment: processing audio
DEBUG:rhasspy3.vad:segment: speaking started
DEBUG:rhasspy3.vad:segment: speaking ended
INFO:faster_whisper_wav2text:Model loaded

This is my configuration.yaml, largely copied from the default, with modifications for testing and voice selection:

programs:

  # -----------
  # Audio input
  # -----------
  mic:

    # apt-get install alsa-utils
    arecord:
      command: |
        arecord -q -D "${device}" -r 16000 -c 1 -f S16_LE -t raw -
      adapter: |
        mic_adapter_raw.py --samples-per-chunk 1024 --rate 16000 --width 2 --channels 1
      template_args:
        device: "default"

    # https://people.csail.mit.edu/hubert/pyaudio/docs/
    pyaudio:
      command: |
        script/events

    # https://python-sounddevice.readthedocs.io
    sounddevice:
      command: |
        script/events

    # apt-get install gstreamer1.0-tools gstreamer1.0-plugins-base
    gstreamer_udp:
      command: |
        gst-launch-1.0 -v udpsrc address=${address} port=${port} ! rawaudioparse use-sink-caps=false format=pcm pcm-format=${format} sample-rate=${rate} num-channels=${channels} ! audioconvert ! audioresample ! volume volume=3.0 ! level ! fdsink fd=1 sync=false
      template_args:
        format: s16le
        rate: 16000
        channels: 1
        address: "0.0.0.0"
        port: 5000
      adapter: |
        mic_adapter_raw.py --samples-per-chunk 1024 --rate 16000 --width 2 --channels 1

    udp_raw:
      command: |
        bin/udp_raw.py --host ${host} --port ${port}
      template_args:
        host: 0.0.0.0
        port: 5000

  # -------------------
  # Wake word detection
  # -------------------
  wake:

    # https://github.com/Picovoice/porcupine
    # Models: see script/list_models
    porcupine1:
      command: |
        .venv/bin/python3 bin/porcupine_stream.py --model "${model}"
      template_args:
        model: "grasshopper_linux.ppn"

    # https://github.com/Kitt-AI/snowboy
    # Models included in share/
    # Custom wake word: https://github.com/rhasspy/snowboy-seasalt
    snowboy:
      command: |
        .venv/bin/python3 bin/snowboy_raw_text.py --model "${model}"
      adapter: |
        wake_adapter_raw.py
      template_args:
        model: "share/snowboy.umdl"

    # https://github.com/mycroftAI/mycroft-precise
    # Model included in share/
    precise-lite:
      command: |
        .venv/bin/python3 bin/precise.py "${model}"
      adapter: |
        wake_adapter_raw.py
      template_args:
        model: "share/hey_mycroft.tflite"

    # TODO: snowman
    # https://github.com/Thalhammer/snowman/

  # ------------------------
  # Voice activity detection
  # ------------------------
  vad:

    # https://github.com/snakers4/silero-vad
    # Model included in share/
    silero:
      command: |
        script/speech_prob "${model}"
      adapter: |
        vad_adapter_raw.py --rate 16000 --width 2 --channels 1 --samples-per-chunk 512
      template_args:
        model: "share/silero_vad.onnx"

    # https://pypi.org/project/webrtcvad/
    webrtcvad:
      command: |
        script/speech_prob ${sensitivity}
      adapter: |
        vad_adapter_raw.py --rate 16000 --width 2 --channels 1 --samples-per-chunk 480
      template_args:
        sensitivity: 3

    # Uses rms energy threshold.
    # For testing only.
    energy:
      command: |
        bin/energy_speech_prob.py --threshold ${threshold} --width 2 --samples-per-chunk 1024
      adapter: |
        vad_adapter_raw.py --rate 16000 --width 2 --channels 1 --samples-per-chunk 1024
      template_args:
        threshold: 300

  # --------------
  # Speech to text
  # --------------
  asr:

    # https://alphacephei.com/vosk/
    # Models: https://alphacephei.com/vosk/models
    vosk:
      command: |
        script/raw2text "${model}"
      adapter: |
        asr_adapter_raw2text.py --rate 16000 --width 2 --channels 1
      template_args:
        model: "${data_dir}/vosk-model-small-en-us-0.15"

    # Run server: asr vosk
    vosk.client:
      command: |
        client_unix_socket.py var/run/vosk.socket

    # https://stt.readthedocs.io
    # Models: https://coqui.ai/models/
    coqui-stt:
      command: |
        script/raw2text "${model}"
      adapter: |
        asr_adapter_raw2text.py --rate 16000 --width 2 --channels 1
      template_args:
        model: "${data_dir}/english_v1.0.0-large-vocab"

    # Run server: asr coqui-stt
    coqui-stt.client:
      command: |
        client_unix_socket.py var/run/coqui-stt.socket

    # https://github.com/cmusphinx/pocketsphinx
    # Models: https://github.com/synesthesiam/voice2json-profiles
    pocketsphinx:
      command: |
        script/raw2text "${model}"
      adapter: |
        asr_adapter_raw2text.py --rate 16000 --width 2 --channels 1
      template_args:
        model: "${data_dir}/en-us_pocketsphinx-cmu"

    # Run server: asr pocketsphinx
    pocketsphinx.client:
      command: |
        client_unix_socket.py var/run/pocketsphinx.socket

    # https://github.com/openai/whisper
    # Models: tiny.en,tiny,base.en,base,small.en,small,medium.en,medium,large-v1,large-v2,large
    # Languages: af,am,ar,as,az,ba,be,bg,bn,bo,br,bs,ca,cs,cy,da,de,el,en,es,et,
    # eu,fa,fi,fo,fr,gl,gu,ha,haw,he,hi,hr,ht,hu,hy,id,is,it,ja,jw,ka,kk,km,kn,
    # ko,la,lb,ln,lo,lt,lv,mg,mi,mk,ml,mn,mr,ms,mt,my,ne,nl,nn,no,oc,pa,pl,ps,
    # pt,ro,ru,sa,sd,si,sk,sl,sn,so,sq,sr,su,sv,sw,ta,te,tg,th,tk,tl,tr,tt,uk,
    # ur,uz,vi,yi,yo,zh
    whisper:
      command: |
        script/wav2text --language ${language} --model-directory "${data_dir}" "${model}" "{wav_file}"
      adapter: |
        asr_adapter_wav2text.py
      template_args:
        language: "en"
        model: "tiny.en"

    # Run server: asr whisper
    whisper.client:
      command: |
        client_unix_socket.py var/run/whisper.socket

    # https://github.com/ggerganov/whisper.cpp/
    # Models: https://huggingface.co/datasets/ggerganov/whisper.cpp
    whisper-cpp:
      command: |
        script/wav2text "${model}" "{wav_file}"
      adapter: |
        asr_adapter_wav2text.py
      template_args:
        model: "${data_dir}/ggml-tiny.en.bin"

    # Run server: asr whisper-cpp
    whisper-cpp.client:
      command: |
        client_unix_socket.py var/run/whisper-cpp.socket

    # https://github.com/guillaumekln/faster-whisper/
    # Models: https://github.com/rhasspy/models/releases/tag/v1.0
    # (asr_faster-whisper-*)
    faster-whisper:
      command: |
        script/wav2text --language ${language} "${model}" "{wav_file}"
      adapter: |
        asr_adapter_wav2text.py
      template_args:
        model: "${data_dir}/tiny-int8"
        language: "en"

    # Run server: asr faster-whisper
    faster-whisper.client:
      command: |
        client_unix_socket.py var/run/faster-whisper.socket

  # --------------
  # Text to speech
  # --------------
  tts:

    # https://github.com/rhasspy/piper/
    # Models: https://github.com/piper/piper/releases/tag/v0.0.2
    piper:
      command: |
        bin/piper --model "${model}" --output_file -
      adapter: |
        tts_adapter_text2wav.py
      template_args:
        model: "${data_dir}/en-us-libritts-high.onnx"
      install:
        command: |
          script/setup.py --destination '${program_dir}/bin'
        check_file: "${program_dir}/bin/piper"
        download:
          command: |
            script/download.py --destination '${data_dir}' '${model}'
        downloads:
          en-us_libritts:
            description: "U.S. English voice"
            check_file: "${data_dir}/en-us-libritts-high.onnx"

    # Run server: tts piper
    piper.client:
      command: |
        client_unix_socket.py var/run/piper.socket

    # https://github.com/rhasspy/larynx/
    # Models: https://rhasspy.github.io/larynx/
    larynx:
      command: |
        .venv/bin/larynx --voices-dir "${data_dir}" --voice "${voice}"
      adapter: |
        tts_adapter_text2wav.py
      template_args:
        voice: "en-us"

    # Run server: tts larynx
    larynx.client:
      command: |
        bin/larynx_client.py ${url} ${voice}
      template_args:
        url: "http://localhost:5002/process"
        voice: "en-us"
      adapter: |
        tts_adapter_text2wav.py

    # https://github.com/espeak-ng/espeak-ng/
    # apt-get install espeak-ng
    espeak-ng:
      command: |
        espeak-ng -v "${voice}" --stdin -w "{temp_file}"
      adapter: |
        tts_adapter_text2wav.py --temp_file
      template_args:
        voice: "en-us"

    # http://www.festvox.org/flite/
    # Models: https://github.com/rhasspy/models/releases/tag/v1.0
    # (tts_flite-*)
    flite:
      command: |
        flite -voice "${voice}" -o "{temp_file}"
      template_args:
        voice: "cmu_us_slt"
      adapter: |
        tts_adapter_text2wav.py --temp_file

    # http://www.cstr.ed.ac.uk/projects/festival/
    # apt-get install festival festival-<lang>
    festival:
      command: |
        text2wave -o "{temp_file}" -eval "(voice_${voice})"
      template_args:
        voice: "cmu_us_slt_arctic_hts"
      adapter: |
        tts_adapter_text2wav.py --temp_file

    # https://tts.readthedocs.io
    # Models: see script/list_models
    coqui-tts:
      command: |
        .venv/bin/tts --model_name "${model}" --out_path "{temp_file}" --text "{text}"
      adapter: |
        tts_adapter_text2wav.py --temp_file --text
      template_args:
        model: "tts_models/en/ljspeech/vits"
        speaker_id: ""

    coqui-tts.client:
      command: |
        tts_adapter_http.py "${url}" --param speaker_id "${speaker_id}"
      template_args:
        url: "http://localhost:5002/api/tts"
        speaker_id: ""

    # http://mary.dfki.de/
    # Models: https://github.com/synesthesiam/opentts/releases/tag/v2.1
    # (marytts-voices.tar.gz)
    marytts:
      command: |
        bin/marytts.py "${url}" "${voice}"
      template_args:
        url: "http://localhost:59125/process"
        voice: "cmu-slt-hsmm"
      adapter: |
        tts_adapter_text2wav.py

    # https://github.com/mycroftAI/mimic3
    # Models: https://mycroftai.github.io/mimic3-voices/
    mimic3:
      command: |
        .venv/bin/mimic3 --voices-dir "${data_dir}" --voice "${voice}" --stdout
      adapter: |
        tts_adapter_text2wav.py
      template_args:
        voice: "apope"

    # Run server: tts mimic3
    mimic3.client:
      command: |
        client_unix_socket.py var/run/mimic3.socket

  # ------------------
  # Intent recognition
  # ------------------
  intent:

    # Simple regex matching
    regex:
      command: |
        bin/regex.py -i TurnOn "turn on (the )?(?P<name>.+)"

    # TODO: fsticuffs
    # https://github.com/rhasspy/rhasspy-nlu

    # TODO: hassil
    # https://github.com/home-assistant/hassil

    # TODO: rapidfuzz
    # https://github.com/rhasspy/rhasspy-fuzzywuzzy

    # TODO: snips-nlu
    # https://snips-nlu.readthedocs.io

  # ---------------
  # Intent Handling
  # ---------------
  handle:

    # Text only: repeats transcript back
    repeat:
      command: |
        cat
      shell: true
      adapter: |
        handle_adapter_text.py

    # Text only: send to HA Assist
    # https://www.home-assistant.io/docs/assist
    #
    # 1. Change server url
    # 2. Put long-lived access token in etc/token
    home_assistant:
      command: |
        bin/converse.py --language "${language}" "${url}" "${token_file}"
      adapter: |
        handle_adapter_text.py
      template_args:
        url: "http://10.0.0.153:8123/api/conversation/process"
        token_file: "${data_dir}/token"
        language: "en"

    # Intent only: answer English date/time requests
    date_time:
      command: |
        bin/date_time.py
      adapter: |
        handle_adapter_text.py

    # Intent only: produces canned response to regex intent system
    test:
      command: |
        name="$(jq -r .slots.name)"
        echo "Turned on ${name}."
      shell: true
      adapter: |
        handle_adapter_json.py

  # ------------
  # Audio output
  # ------------
  snd:

    # apt-get install alsa-utils
    aplay:
      command: |
        aplay -q -D "${device}" -r 22050 -f S16_LE -c 1 -t raw
      adapter: |
        snd_adapter_raw.py --rate 22050 --width 2 --channels 1
      template_args:
        device: "default"

    udp_raw:
      command: |
        bin/udp_raw.py --host "${host}" --port ${port}
      template_args:
        host: "127.0.0.1"
        port: 5001

  # -------------------
  # Remote base station
  # -------------------
  remote:

    # Sample tool to communicate with websocket API.
    # Use rhasspy3/bin/satellite_run.py
    websocket:
      command: |
        script/run "${uri}"
      template_args:
        uri: "ws://localhost:13331/pipeline/asr-tts"

# -----------------------------------------------------------------------------

servers:
  asr:
    vosk:
      command: |
        script/server "${model}"
      template_args:
        model: "${data_dir}/vosk-model-small-en-us-0.15"

    coqui-stt:
      command: |
        script/server "${model}"
      template_args:
        model: "${data_dir}/english_v1.0.0-large-vocab"

    pocketsphinx:
      command: |
        script/server "${model}"
      template_args:
        model: "${data_dir}/en-us_pocketsphinx-cmu"

    whisper:
      command: |
        script/server --model-directory "${data_dir}" --language ${language} --device ${device} ${model}
      template_args:
        language: "en"
        model: "tiny.en"
        device: "cpu"  # or cuda

    whisper-cpp:
      command: |
        script/server "${model}"
      template_args:
        model: "${data_dir}/ggml-tiny.en.bin"

    faster-whisper:
      command: |
        script/server --language ${language} --device ${device} "${model}"
      template_args:
        language: "en"
        model: "${data_dir}/tiny-int8"
        device: "cpu"  # or cuda

  tts:
    mimic3:
      command: |
        script/server --voice "${voice}" "${data_dir}"
      template_args:
        voice: "en_US/ljspeech_low"

    piper:
      command: |
        script/server "${model}"
      template_args:
        model: "${data_dir}/en-us-libritts-high.onnx"

    larynx:
      command: |
        script/server --voices-dir "${data_dir}" --host "${host}"
      template_args:
        host: "127.0.0.1"

    coqui-tts:
      command: |
        script/server

# -----------------------------------------------------------------------------

# Example satellites
# satellites:
#   default:
#     mic:
#       name: arecord
#     wake:
#       name: porcupine1
#     remote:
#       name: websocket
#     snd:
#       name: aplay

# -----------------------------------------------------------------------------

# Example pipelines
pipelines:

  # English (default)
  default:
    mic:
      name: arecord
    wake:
      name: porcupine1
    vad:
      name: silero
    asr:
      name: faster-whisper
    # intent:
    #   name: regex
    handle:
      name: repeat
    #  name: home_assistant
    tts:
      name: piper
    snd:
      name: aplay

  # # German
  # de:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: de
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/de-thorsten-low.onnx"

  # # French
  # fr:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: fr
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/fr-siwis-low.onnx"

  # # Spanish
  # es:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: es
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/es-carlfm-low.onnx"

  # # Italian
  # it:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: it
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/it-riccardo_fasol-low.onnx"

  # # Catalan
  # ca:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: ca
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/ca-upc_ona-low.onnx"

  # # Danish
  # da:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: da
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/da-nst_talesyntese-medium.onnx"

  # # Dutch
  # nl:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: nl
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/nl-nathalie-low.onnx"

  # # Norwegian
  # no:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: no
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/no-talesyntese-medium.onnx"

  # # Ukrainian
  # uk:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: uk
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/uk-lada-low.onnx"

  # # Vietnamese
  # vi:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: vi
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/vi-vivos-low.onnx"

  # # Chinese
  # zh:
  #   inherit: default
  #   asr:
  #     name: faster-whisper
  #     template_args:
  #       language: zh
  #   tts:
  #     name: piper
  #     template_args:
  #       model: "${data_dir}/zh-cn-huayan-low.onnx"

[Shulyaka]

Hi!

Please try this fix: #30.

Probably duplicate of #29.

[Shulyaka]

By the way, you don't need to copy the default configuration - only overwrite the changed parts. Both configurations are used.

[ethereal-engineer]

By the way, you don't need to copy the default configuration - only overwrite the changed parts. Both configurations are used.

Oh sweet. I kind of thought that would be the case but I was hacking around with it and seeing what worked, so had copied the entire lot across so I could play with all the options.

[ethereal-engineer]

Well, after applying that fix and running it for several minutes of testing, that appears to have done the trick. I'm an ex-dev but Python is one of the few languages I haven't used. I'm glad for the moment that I didn't need to dig deeper on this personally. Thank you for noticing my issue and for your fix.