> Let's build a webapp that allows the user to choose an audio file, we slice it into timestamped word regions with transformers.js + onnx-community/whisper-base_timestamped (with language selection). The user can then scroll through the waveform (that shows the found word regions), synchronized with the transcript, and select word regions in the transcript or in the waveform, and fine-tune the selection if they need to, and then export the slice as WAV.

and see what pops out.

* the web site's layout (eyebrows, glows, em-dashes, middots, that certain breathless tone in the pricing box) (edit: see the source for https://vocalslice.com too)

* the changelog listing, eg. the fact that you can search by language code as a headline feature

* in the app, things like "Click here to choose an audio file to transcribe – swap it any time" are Claude-like copy; I'd be surprised if you _couldn't_ change the audio file at any time?

* looking at https://vocalslice.com/demo.webp, another claudeism is the glowy status indicator blob, it really likes to add that by default.

If OP gets a business going with this, more power to them and all the best, but it really is something that's buildable pretty easily with transformers.js and models like onnx-community/whisper-small_timestamped. (Speaking of which, I hope OP is hosting the models somewhere themselves and not trusting, say, Hugging Face with the continuity of their business.)

Finally decided to do something about it: https://github.com/jdan/98.css/pull/230

Is there any point for (new) .bz2 archives in the era of Zstd?

    >>> 'x\u2009   a'.split()
    ['x', 'a']
    # incorrect; in bytes mode, `\S` doesn't know about unicode whitespace
    >>> list(re.finditer(br'\S+', 'x\u2009   a'.encode()))
    [, ]
    # correct, in unicode mode
    >>> list(re.finditer(r'\S+', 'x\u2009   a'))
    [, ]

All too often, new models' codebases are just a dump of code that installs half the universe in dependencies for no reason, etc.

I'd imagine not, since `ruff format` and `ruff check` are separate things too.

For instance, try adding `np.random.shuffle(ref_s[0])` after the line `ref_s = self.voices[voice]`...

EDIT: be careful with your system volume settings if you do this.

https://github.com/espeak-ng/espeak-ng/blob/a4ca101c99de3534...

https://github.com/KittenML/KittenTTS/pull/21 https://github.com/KittenML/KittenTTS/pull/24 https://github.com/KittenML/KittenTTS/pull/25

If you have `uv` installed, you can try my merged ref that has all of these PRs (and #22, a fix for short generation being trimmed unnecessarily) with

    uvx --from git+https://github.com/akx/KittenTTS.git@pr-21-22-24-25 kittentts --output output.wav --text "This high quality TTS model works without a GPU"