Check memory/running of onnx is not cold

[willwade]

its still slow to run onnx tts. 6 seconds on average. I think the onnx models cannot be sitting in memory.. its too slow

[willwade]

here are some test sentences. English

What do you think about?
Tell me about?
I've been think about?
Did you know that?
Did you see?
I would like to?
I think that?
I've been meaning to tell you
By the way
I'd rather
Someday I want to
Topic continuations
And then...
Topic continuations
Keep going.
Topic continuations
Right.
Topic continuations
Oh yeah.
How are you? Are you Ok?
I agree.

Traditional Chinese

你覺得怎麼樣？
告訴我一下？
我有想過嗎？
你知道嗎？
你有沒有看到？
我想？
我認為是這樣嗎？
我一直想告訴你
順便一提
我寧願
有一天我想要
主題延續
進而...
主題延續
繼續前進。
主題延續
正確的。
主題延續
哦，是的。
你好嗎？你還好嗎？
我同意。

Kurdish Sorani```

ڕای ئێوە چییە؟
دەربارەی؟
بیرم لێکردۆتەوە؟
ئایا ئەوەت زانیوە؟
بینیوتە؟
حەز دەکەم؟
پێم وایە ئەوە؟
من مەبەستم بووە پێت بڵێم
بە ڕاست
من پێم باشە
ڕۆژێک دێت دەمەوێت
بەردەوامی بابەتەکان
و پاشان...
بەردەوامی بابەتەکان
بەردەوام بە.
بەردەوامی بابەتەکان
ڕاست.
بەردەوامی بابەتەکان
ئای بەڵێ.
چۆنی؟ ئایا تۆ باشیت؟
منیش لەگەڵ ئەوەدام.

[acerjay]

upon reviewing the code, around tts_wrapper\engines\sherpaonnx\client.py: the 6 second render time. For a small sample text above, PyAudio writes it around 2 second while sherpaaonnx takes 4 seconds to generate tts audio. I'll try some approach to improve the performance of this function.

[willwade]

Interesting. I’ve been wondering for a while if my approach at playing generated bytes is fast or could be faster (in tts wrapper). Would love a better approach.

Even 2seconds is too slow to write.