todo

rust-whisper-baked-lib/src/lib.rs

@@ -25,7 +25,6 @@ pub fn wav_channel<F>(
 ) where F: FnMut(Result<rust_whisper_lib::Transcribed, String>) + Send + 'static {
    flags.model_path = None;
    flags.model_buffer = Some(include_bytes!("../../models/ggml-distil-medium.en.bin").to_vec());
-   flags.model_buffer = Some(include_bytes!("../../models/ggml-base.en.bin").to_vec());
    rust_whisper_lib::wav_channel(flags, handler_fn);
 }
 

rust-whisper-baked/src/main.rs

@@ -99,6 +99,7 @@ impl Destutterer {
 
       let next_words = Words::from_string(next.clone());
       let mut n = self.prev.to_comparable_words().len().clamp(0, next_words.to_comparable_words().len());
+      //println!("n={} prev='{:?}' next='{:?}'", n, self.prev.to_comparable_words(), next_words.to_comparable_words());
       while n > 0 {
          let (prev_s, _) = self.prev.last_n_comparable_to_string(n);
          let (next_s, next_idx) = next_words.first_n_comparable_to_string(n);

todo.yaml

@@ -1,7 +1,9 @@
 todo:
-- overlap without stop words
+- split on silence-ish instead of duration
 - rust-whisper warn when transcription time ~ input time
 scheduled: []
 done:
 - todo: need to overlap without ANY puctuation, which i can do by breaking into words
   ts: Tue Jan  2 18:23:00 MST 2024
+- todo: overlap without stop words
+  ts: Wed Jan  3 08:22:14 MST 2024