From 11384921323f7b08271be0607d8df774489a1f7d Mon Sep 17 00:00:00 2001 From: historia Date: Wed, 19 Aug 2026 02:10:13 -0400 Subject: fix: reduce chunk size to 40 --- README.md | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'README.md') diff --git a/README.md b/README.md index fd24a8d..0913632 100644 --- a/README.md +++ b/README.md @@ -188,6 +188,10 @@ Third-party wheels: https://mjunya.com/flash-attention-prebuild-wheels/ (hosted Transcription affects the output a lot. Whisper is okay, but does not give perfect transcription. A manual transcription passed via `--transcription` is better. +Keep `CHUNK_SIZE_WORDS` small (default 40). Every API call is a single model generation: long generations lose prosody, can degrade into garbled audio, and text past the model's token limit is never spoken. If parts of a book sound flat, monotone, or garbled, the chunk size is the first thing to check. + +`MIN_DELAY_BETWEEN_CHUNKS` only matters for hosted demos (rate limits); a local server needs no delay (default 0). + Manual transcription, imperfect whisper transcription, and `--no-transcription` each provide different results. Usually the most accurate transcription is the best, but sometimes `--no-transcription` can produce a flat tone that might be preferable for certain voices. Even tiny amounts of pause between phrases in the sample audio can have a big impact. Try increasing or decreasing them. -- cgit v1.2.3