X-vector based TTS model packaging broken in tts.sh

[G-Thor]

Describe the bug
PR #5579 broke xvector-conditioned TTS model packaging. In stage 9 of tts.sh, spk_xvector.ark was replaced with {spk_embed_tag}.ark, which in my recipe resolves to xvector.ark. That file does not exist whereas spk_xvector.ark does.

Basic environments:

• OS information: Linux 4.18.0-513.18.1.el8_9.x86_64 Updated sphinx documents #1 SMP Wed Feb 21 21:34:36 UTC 2024 x86_64
• python version: 3.9.18 (main, Sep 11 2023, 13:41:44) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 2.1.0
• Git hash: d0047402e830a3c53e8b590064af4bf70415fb3b
  • Commit date: Mon Mar 4 22:19:02 2024 +0000
• pytorch version [e.g. pytorch 1.4.0]

Task information:

• Task: TTS
• Recipe: talromur (but applies to all)
• ESPnet2

To Reproduce
Steps to reproduce the behavior:

1. run any xvector-based recipe up until the model packaging stage (e.g. jtubespeech)

• e.g. cd egs2/jtubespeech/tts1; ./run.sh --stop-stage 8

3. execute ./run.sh --stage 9 --stop-stage 9
4. Observe command output

To Fix
This error originates in the following lines, and can be fixed by modifying lines 1133 and 1134 of tts.sh, changing {spk_embed_tag}.ark to spk_{spk_embed_tag}.ark and {spk_embed_tag}.scp to spk_{spk_embed_tag}.scp :

espnet/egs2/TEMPLATE/tts1/tts.sh

Lines 1131 to 1135 in ca7716f

	if "${use_spk_embed}"; then
	for dset in "${train_set}" ${test_sets}; do
	_opts+=" --option ${dumpdir}/${spk_embed_tag}/${dset}/${spk_embed_tag}.scp"
	_opts+=" --option ${dumpdir}/${spk_embed_tag}/${dset}/${spk_embed_tag}.ark"
	done

I'm not sure whether or how this may affect the new speaker embedding implementation, perhaps the PR author @ftshijt has insight into that?
By the way, thanks for the great work on better integrating speaker embeddings into TTS recipes. I look forward to training an Icelandic speaker embedding model for multi-speaker TTS.

[ftshijt]

Thanks for the note! You are correct, the packing should be updated. I will have a check tomorrow and make a PR soon (hopefully also to upload the pre-trained TTS model on the new speaker embedding at the same time)