ch 16: PortAudio library not found

[dblanding]

I now have Raspberry Pi OS Full (64-bit) and have successfully installed piper-tts.
I have progressed to Ch16, p8 where I ssh to the robot:

doug@raspi4:~ $ robotpython
Python 3.13.5 (main, Jun 25 2025, 18:55:22) [GCC 14.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from robot.common.speaking import speak
2025-10-13 06:09:19.400186934 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    from robot.common.speaking import speak
  File "/home/doug/robot/common/speaking.py", line 3, in <module>
    import sounddevice as sd
  File "/home/doug/robotvenv/lib/python3.13/site-packages/sounddevice.py", line 71, in <module>
    raise OSError('PortAudio library not found')
OSError: PortAudio library not found

[dblanding]

I tried adding this into the deploy_speaking file (see below) and then tried again:

doug@raspi4:~ $ robotpython
Python 3.13.5 (main, Jun 25 2025, 18:55:22) [GCC 14.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from robot.common.speaking import speak
2025-10-13 06:54:19.138070729 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"

deploy_speaking.py

[dblanding]

My deploy_speaking.py still has an apt install clause for pulseaudio which the Lite OS needed, but which may have disrupted the audio libraries already installed in the Full version. Removing this apt install clause in my deploy_speaking.py file doesn't have any effect now.

[dannystaple]

Hmm, I am seeing these event at page 6:

danny@learnrob3-2:~ $ robotpython -m piper -m en_US-ryan-low "Hello there."
2025-10-13 19:11:50.821777391 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"

paplay and parecord definitely worked here with no parameters, so it's not the audio device. But it's now asking for a GPU - it was not before.

[dannystaple]

Does that GPU device exist on the prior SD card?

Trying SD #0 - bookworm, Raspberry Pi OS lite. The file /sys/class/drm/card1/device/vendor does not exist (nor for card 0). piper version - piper-tts==1.3.0. Python tags include aarch64. Piper works fine here.

[dannystaple]

On SD Card No 1, Bookwarm, aarch64. Piper works and speaks. The /sys/class/drm/card1/device/vendor also doesn't exist there. Also piper-tts==1.3.0.

So what about that trixie configuration is different? It's not the piper version, it's also not the audio config - but it's trying to load for a GPU for some reason I think. That part W:onnxruntime:Default, device_discovery.cc:164 of the error suggests its the neural network runtime.

[dannystaple]

Dig a bit of digging and got to https://onnxruntime.ai/ - the underlying neural network engine runtime powering this. I'm able to reproduce in a much smaller area:

danny@learnrob3-2:~ $ robotpython -
Python 3.13.5 (main, Jun 25 2025, 18:55:22) [GCC 14.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import onnxruntime as ort
2025-10-13 19:40:20.052020623 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"

I can now look for clues in the onnxruntime repo.

[dannystaple]

A simple thing to check for - what is the package version of onnxruntime, and how is that version mandated by piper-tts? This line - https://github.com/OHF-Voice/piper1-gpl/blob/main/setup.py#L56C13-L56C26 - which is onnx, any version between 1 and 2.
I'll check the other SD cards...

Sd card 2 - onnxruntime==1.23.1

[dannystaple]

SD card 0 - onnxruntime==1.19.2 - that is older.
SD Card 1 - ``onnxruntime==1.22.1` - one minor version behind.
Let's try a pip update on sd card 1...
The error message is reproduced... But...
I also heard the "hello there" from the headset eventually.

danny@learnrob3-1:~ $ robotpython -m piper -m en_US-ryan-low "Hello there."
2025-10-13 19:51:28.586451797 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"
danny@learnrob3-1:~ $ echo $?
0

This particular message about drm might be a red herring. Trying again on sd card 2.

[dannystaple]

Trying a longer phrase on SD card 2 - this is interesting:

danny@learnrob3-2:~ $ robotpython -m piper -m en_US-ryan-low "Hello there, I did say something but it took a while."
2025-10-13 19:56:01.173419807 [W:onnxruntime:Default, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/home/danny/robotvenv/lib/python3.13/site-packages/piper/__main__.py", line 230, in <module>
    main()
    ~~~~^^
  File "/home/danny/robotvenv/lib/python3.13/site-packages/piper/__main__.py", line 213, in main
    player.play(audio_chunk.audio_int16_bytes)
    ~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/danny/robotvenv/lib/python3.13/site-packages/piper/audio_playback.py", line 52, in play
    self._proc.stdin.write(audio_bytes)
    ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
BrokenPipeError: [Errno 32] Broken pipe
danny@learnrob3-2:~ $ echo $?
1

I do not hear anything, and there is an audio playback error. The GPU thing may still be a red herring. I could force that out of the equation, perhaps by pip installing the earlier onnxruntime.
robotpython -mpip install onnxruntime==1.22.1
Let's see if I then get audio...
I now get only the player broken pipe error:

$ robotpython -m piper -m en_US-ryan-low "Hello there, I did say something but it took a while."
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/home/danny/robotvenv/lib/python3.13/site-packages/piper/__main__.py", line 230, in <module>
    main()
    ~~~~^^
  File "/home/danny/robotvenv/lib/python3.13/site-packages/piper/__main__.py", line 213, in main
    player.play(audio_chunk.audio_int16_bytes)
    ~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/danny/robotvenv/lib/python3.13/site-packages/piper/audio_playback.py", line 52, in play
    self._proc.stdin.write(audio_bytes)
    ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
BrokenPipeError: [Errno 32] Broken pipe

I think either I'm seeing a new error, or I've ruled out that drm file as a red herring here. @dblanding do you get different behavior if you try: robotpython -mpip install onnxruntime==1.22.1 then run a piper voice command?

[dblanding]

First of all, I made a comment on 10/11/25 about the way robotpython -m piper -m en_US-ryan-low "Hello there." works. I noted that it creates the file output.wav, which can then be played with the command paplay output.wav.
Now, I will try the command robotpython -mpip install onnxruntime==1.22.1. Here's what I get:

doug@raspi4:~ $ robotpython -mpip install onnxruntime==1.22.1
Collecting onnxruntime==1.22.1
  Downloading onnxruntime-1.22.1-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl.metadata (4.9 kB)
Requirement already satisfied: coloredlogs in ./robotvenv/lib/python3.13/site-packages (from onnxruntime==1.22.1) (15.0.1)
Requirement already satisfied: flatbuffers in ./robotvenv/lib/python3.13/site-packages (from onnxruntime==1.22.1) (25.9.23)
Requirement already satisfied: numpy>=1.21.6 in /usr/lib/python3/dist-packages (from onnxruntime==1.22.1) (2.2.4)
Requirement already satisfied: packaging in /usr/lib/python3/dist-packages (from onnxruntime==1.22.1) (25.0)
Requirement already satisfied: protobuf in ./robotvenv/lib/python3.13/site-packages (from onnxruntime==1.22.1) (6.32.1)
Requirement already satisfied: sympy in ./robotvenv/lib/python3.13/site-packages (from onnxruntime==1.22.1) (1.14.0)
Requirement already satisfied: humanfriendly>=9.1 in ./robotvenv/lib/python3.13/site-packages (from coloredlogs->onnxruntime==1.22.1) (10.0)
Requirement already satisfied: mpmath<1.4,>=1.1.0 in ./robotvenv/lib/python3.13/site-packages (from sympy->onnxruntime==1.22.1) (1.3.0)
Downloading onnxruntime-1.22.1-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl (14.5 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 14.5/14.5 MB 406.0 kB/s eta 0:00:00
Installing collected packages: onnxruntime
  Attempting uninstall: onnxruntime
    Found existing installation: onnxruntime 1.23.1
    Uninstalling onnxruntime-1.23.1:
      Successfully uninstalled onnxruntime-1.23.1
Successfully installed onnxruntime-1.22.1

[dblanding]

I went back to the interactive prompt and now it works.

doug@raspi4:~ $ robotpython
Python 3.13.5 (main, Jun 25 2025, 18:55:22) [GCC 14.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from robot.common.speaking import speak
>>> speak("hello there!")
>>> speak("Lets make robots.")
>>> 

[dblanding]

>>> speak("Nice job, Danny!")

[dannystaple]

Going to keep this open for now - as I probably want to see why downgrading that package helps.