Merge pull request #174 from thewh1teagle/feat/improve-basic-example

feat: improve basic example

.gitignore

@@ -1,4 +1,6 @@
 **/target
 **/Cargo.lock
 /.idea
-/.vscode
+/.vscode
+*.bin
+*.wav

examples/basic_use.rs

@@ -1,54 +1,59 @@
-#![allow(clippy::uninlined_format_args)]
+/*
+wget https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.bin
+wget https://github.com/ggerganov/whisper.cpp/raw/master/samples/jfk.wav
+cargo run --example basic_use ggml-tiny.bin jfk.wav
+*/
 
 use whisper_rs::{FullParams, SamplingStrategy, WhisperContext, WhisperContextParameters};
 
-// note that running this example will not do anything, as it is just a
+fn main() {
-// demonstration of how to use the library, and actual usage requires
+    let model_path = std::env::args()
-// more dependencies than the base library.
+        .nth(1)
-pub fn usage() -> Result<(), &'static str> {
+        .expect("Please specify path to model");
+    let wav_path = std::env::args()
+        .nth(2)
+        .expect("Please specify path to wav file");
+    let language = "en";
+
+    let samples: Vec<i16> = hound::WavReader::open(wav_path)
+        .unwrap()
+        .into_samples::<i16>()
+        .map(|x| x.unwrap())
+        .collect();
+
     // load a context and model
-    let ctx = WhisperContext::new_with_params("path/to/model", WhisperContextParameters::default())
+    let ctx = WhisperContext::new_with_params(&model_path, WhisperContextParameters::default())
         .expect("failed to load model");
-    // make a state
+
     let mut state = ctx.create_state().expect("failed to create state");
 
-    // create a params object
-    // note that currently the only implemented strategy is Greedy, BeamSearch is a WIP
-    // n_past defaults to 0
     let mut params = FullParams::new(SamplingStrategy::Greedy { best_of: 1 });
 
-    // edit things as needed
-    // here we set the number of threads to use to 1
-    params.set_n_threads(1);
-    // we also enable translation
-    params.set_translate(true);
     // and set the language to translate to to english
-    params.set_language(Some("en"));
+    params.set_language(Some(&language));
+
     // we also explicitly disable anything that prints to stdout
     params.set_print_special(false);
     params.set_print_progress(false);
     params.set_print_realtime(false);
     params.set_print_timestamps(false);
 
-    // assume we have a buffer of audio data
-    // here we'll make a fake one, integer samples, 16 bit, 16KHz, stereo
-    let audio_data = vec![0_i16; 16000 * 2];
-
     // we must convert to 16KHz mono f32 samples for the model
     // some utilities exist for this
     // note that you don't need to use these, you can do it yourself or any other way you want
     // these are just provided for convenience
     // SIMD variants of these functions are also available, but only on nightly Rust: see the docs
-    let mut inter_audio_data = Vec::with_capacity(audio_data.len());
+    let mut inter_samples = vec![Default::default(); samples.len()];
-    whisper_rs::convert_integer_to_float_audio(&audio_data, &mut inter_audio_data)
+
+    whisper_rs::convert_integer_to_float_audio(&samples, &mut inter_samples)
         .expect("failed to convert audio data");
-    let audio_data = whisper_rs::convert_stereo_to_mono_audio(&inter_audio_data)
+    let samples = whisper_rs::convert_stereo_to_mono_audio(&inter_samples)
         .expect("failed to convert audio data");
 
     // now we can run the model
     // note the key we use here is the one we created above
     state
-        .full(params, &audio_data[..])
+        .full(params, &samples[..])
         .expect("failed to run model");
 
     // fetch the results
@@ -67,10 +72,4 @@ pub fn usage() -> Result<(), &'static str> {
             .expect("failed to get segment end timestamp");
         println!("[{} - {}]: {}", start_timestamp, end_timestamp, segment);
     }
-
-    Ok(())
-}
-
-fn main() {
-    println!("running this example does nothing! see the source code for usage");
 }