Merge pull request #36 from shivenmian/u/shivenmian/fix_local

fix: add back whisper-rust local scripts, compile binary in start script

.gitignore

@@ -168,3 +168,4 @@ cython_debug/
 # ignore the aifs index files
 _.aifs
 01OS/output_audio.wav
+.DS_Store

01OS/.env.example

@@ -5,6 +5,7 @@
 # else we use whisper.cpp and piper local models
 ALL_LOCAL=False
 WHISPER_MODEL_NAME="ggml-tiny.en.bin"
+WHISPER_MODEL_URL="https://huggingface.co/ggerganov/whisper.cpp/resolve/main/"
 
 # Uncomment to set your OpenAI API key
 # OPENAI_API_KEY=sk-...

01OS/01OS/server/stt/stt.py

@@ -57,7 +57,7 @@ def run_command(command):
 
 def get_transcription_file(wav_file_path: str):
     local_path = os.path.join(os.path.dirname(__file__), 'local_service')
-    whisper_rust_path = os.path.join(local_path, 'whisper-rust')
+    whisper_rust_path = os.path.join(os.path.dirname(__file__), 'whisper-rust', 'target', 'release')
     model_name = os.getenv('WHISPER_MODEL_NAME')
     if not model_name:
         raise EnvironmentError("WHISPER_MODEL_NAME environment variable is not set.")

01OS/01OS/server/stt/whisper-rust/.gitignore

@@ -0,0 +1,10 @@
+# Generated by Cargo
+# will have compiled files and executables
+debug/
+target/
+
+# These are backup files generated by rustfmt
+**/*.rs.bk
+
+# MSVC Windows builds of rustc generate these, which store debugging information
+*.pdb

01OS/01OS/server/stt/whisper-rust/Cargo.toml

@@ -0,0 +1,14 @@
+[package]
+name = "whisper-rust"
+version = "0.1.0"
+edition = "2021"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+anyhow = "1.0.79"
+clap = { version = "4.4.18", features = ["derive"] }
+cpal = "0.15.2"
+hound = "3.5.1"
+whisper-rs = "0.10.0"
+whisper-rs-sys = "0.8.0"

01OS/01OS/server/stt/whisper-rust/src/main.rs

@@ -0,0 +1,34 @@
+mod transcribe;
+
+use clap::Parser;
+use std::path::PathBuf;
+use transcribe::transcribe;
+
+#[derive(Parser, Debug)]
+#[command(author, version, about, long_about = None)]
+struct Args {
+    /// This is the model for Whisper STT
+    #[arg(short, long, value_parser, required = true)]
+    model_path: PathBuf,
+    
+    /// This is the wav audio file that will be converted from speech to text
+    #[arg(short, long, value_parser, required = true)]
+    file_path: Option<PathBuf>,
+}
+
+fn main() {
+
+    let args = Args::parse();
+
+    let file_path = match args.file_path {
+        Some(fp) => fp,
+        None => panic!("No file path provided")
+    };
+
+    let result = transcribe(&args.model_path, &file_path);
+
+    match result {
+        Ok(transcription) => print!("{}", transcription),
+        Err(e) => panic!("Error: {}", e),
+    }
+}

01OS/01OS/server/stt/whisper-rust/src/transcribe.rs

@@ -0,0 +1,64 @@
+use whisper_rs::{FullParams, SamplingStrategy, WhisperContext, WhisperContextParameters};
+use std::path::PathBuf;
+
+
+/// Transcribes the given audio file using the whisper-rs library.
+///
+/// # Arguments
+/// * `model_path` - Path to Whisper model file
+/// * `file_path` - A string slice that holds the path to the audio file to be transcribed.
+///
+/// # Returns
+///
+/// A Result containing a String with the transcription if successful, or an error message if not.
+pub fn transcribe(model_path: &PathBuf, file_path: &PathBuf) -> Result<String, String> {
+
+    let model_path_str = model_path.to_str().expect("Not valid model path");
+    // Load a context and model
+    let ctx = WhisperContext::new_with_params(
+        model_path_str, // Replace with the actual path to the model
+        WhisperContextParameters::default(),
+    )
+    .map_err(|_| "failed to load model")?;
+
+    // Create a state
+    let mut state = ctx.create_state().map_err(|_| "failed to create state")?;
+
+    // Create a params object
+    // Note that currently the only implemented strategy is Greedy, BeamSearch is a WIP
+    let mut params = FullParams::new(SamplingStrategy::Greedy { best_of: 1 });
+
+    // Edit parameters as needed
+    params.set_n_threads(1); // Set the number of threads to use
+    params.set_translate(true); // Enable translation
+    params.set_language(Some("en")); // Set the language to translate to English
+    // Disable printing to stdout
+    params.set_print_special(false);
+    params.set_print_progress(false);
+    params.set_print_realtime(false);
+    params.set_print_timestamps(false);
+
+    // Load the audio file
+    let audio_data = std::fs::read(file_path)
+        .map_err(|e| format!("failed to read audio file: {}", e))?
+        .chunks_exact(2)
+        .map(|chunk| i16::from_ne_bytes([chunk[0], chunk[1]]))
+        .collect::<Vec<i16>>();
+
+    // Convert the audio data to the required format (16KHz mono i16 samples)
+    let audio_data = whisper_rs::convert_integer_to_float_audio(&audio_data);
+
+    // Run the model
+    state.full(params, &audio_data[..]).map_err(|_| "failed to run model")?;
+
+    // Fetch the results
+    let num_segments = state.full_n_segments().map_err(|_| "failed to get number of segments")?;
+    let mut transcription = String::new();
+    for i in 0..num_segments {
+        let segment = state.full_get_segment_text(i).map_err(|_| "failed to get segment")?;
+        transcription.push_str(&segment);
+        transcription.push('\n');
+    }
+
+    Ok(transcription)
+}

01OS/start.sh

@@ -71,19 +71,39 @@ if [[ "$ALL_LOCAL" == "True" ]]; then
 
     ## WHISPER
     
-    WHISPER_MODEL_URL="https://huggingface.co/ggerganov/whisper.cpp/resolve/main/"
+    CWD=$(pwd)
-    WHISPER_PATH="$SCRIPT_DIR/01OS/server/stt/local_service"
+
-    if [[ ! -f "${WHISPER_PATH}/${WHISPER_MODEL_NAME}" ]]; then
+    STT_PATH="$SCRIPT_DIR/01OS/server/stt"
-        mkdir -p "${WHISPER_PATH}"
+    WHISPER_RUST_PATH="${STT_PATH}/whisper-rust"
-        curl -L "${WHISPER_MODEL_URL}${WHISPER_MODEL_NAME}" -o "${WHISPER_PATH}/${WHISPER_MODEL_NAME}"
+    cd ${WHISPER_RUST_PATH}
+
+    # Check if whisper-rust executable exists
+    if [[ ! -f "${WHISPER_RUST_PATH}/target/release/whisper-rust" ]]; then
+
+        # Check if Rust is installed. Needed to build whisper executable
+        if ! command -v rustc &> /dev/null; then
+            echo "Rust is not installed or is not in system PATH. Please install Rust before proceeding."
+            exit 1
+        fi
+
+        # Build the Whisper Rust executable
+        cargo build --release
+    fi
+
+    WHISPER_MODEL_PATH="${STT_PATH}/local_service"
+    if [[ ! -f "${WHISPER_MODEL_PATH}/${WHISPER_MODEL_NAME}" ]]; then
+        mkdir -p "${WHISPER_MODEL_PATH}"
+        curl -L "${WHISPER_MODEL_URL}${WHISPER_MODEL_NAME}" -o "${WHISPER_MODEL_PATH}/${WHISPER_MODEL_NAME}"
     fi
 
+    cd $CWD
+
     ## PIPER
 
-    PIPER_FILE_PATH="$SCRIPT_DIR/01OS/server/tts/local_service${PIPER_URL}${PIPER_ASSETNAME}"
+    PIPER_FOLDER_PATH="$SCRIPT_DIR/01OS/server/tts/local_service"
-    if [[ ! -f "$PIPER_FILE_PATH" ]]; then   
+    if [[ ! -f "$PIPER_FOLDER_PATH" ]]; then   
 
-        mkdir -p "${PIPER_FILE_PATH}"
+        mkdir -p "${PIPER_FOLDER_PATH}"
 
         OS=$(uname -s)
         ARCH=$(uname -m)
@@ -104,7 +124,7 @@ if [[ "$ALL_LOCAL" == "True" ]]; then
         CWD=$(pwd)
 
         # Navigate to SCRIPT_DIR/01OS/server/tts/local_service
-        cd $SCRIPT_DIR/01OS/server/tts/local_service
+        cd ${PIPER_FOLDER_PATH}
 
         curl -L "${PIPER_URL}${PIPER_ASSETNAME}" -o "${PIPER_ASSETNAME}"
         tar -xvzf $PIPER_ASSETNAME

README.md

@@ -20,6 +20,8 @@ brew install portaudio ffmpeg
 sudo apt-get install portaudio19-dev ffmpeg
 ```
 
+If you want to run local speech-to-text using Whisper, install Rust. Follow the instructions given [here](https://www.rust-lang.org/tools/install).
+
 ## Setup for usage (experimental):
 
 ```bash