InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid input name: sid

[graybrandonh]

I have been attempting to use piper with the python-tts method and it seems only the following English voice files work:

• en_US-arctic-medium
• en_US-l2arctic-medium
• en_US-libritts-high
• en_US-libritts_r-medium
• en_GB-aru-medium
• en_GB-semaine-medium
• en_GB-vctk-medium

The rest of them throw the following exception.

Traceback (most recent call last):
File "/usr/local/bin/piper", line 8, in
sys.exit(main())
File "/usr/local/lib/python3.10/dist-packages/piper/main.py", line 155, in main
voice.synthesize(text, wav_file, **synthesize_args)
File "/usr/local/lib/python3.10/dist-packages/piper/voice.py", line 95, in synthesize
for audio_bytes in self.synthesize_stream_raw(
File "/usr/local/lib/python3.10/dist-packages/piper/voice.py", line 123, in synthesize_stream_raw
yield self.synthesize_ids_to_raw(
File "/usr/local/lib/python3.10/dist-packages/piper/voice.py", line 166, in synthesize_ids_to_raw
audio = self.session.run(
File "/usr/local/lib/python3.10/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 266, in run
return self._sess.run(output_names, input_feed, run_options)
onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid input name: sid

Any ideas on how to get the other voice files working?

[yosmont]

Same here with the en_US-amy-medium one on a raspberry pi 4.
I've tried the echo 'Welcome to the world of speech synthesis!' | piper --model en_US-amy-medium --output_file welcome.wav cli to try the model with just piper, and it worked

[p0n1]

I'll debug this issue.

[KitB]

You can fix it by commenting out a couple of lines here:

epub_to_audiobook/audiobook_generator/tts_providers/piper_tts_provider.py

Lines 49 to 50 in 8803f42

	"--speaker",
	str(self.config.piper_speaker),

This is not a proper fix and I'd guess will break things for the models that do have multiple speakers (it appears that the affected models are the single-speaker models), but it can make the single-speaker models work in a pinch.