DS-TRANSCRIBER

Need an Offline Speech To Text converter?

Records your mic, and returns a String containing what was said.

Features

• Begins transcription after a long enough pause has been detected
• Change the length of the pause required to begin transcription
• Shows you the audio levels of what is being recorded so that you can...
• Change the audio level of what you deem as silence in your environment

Ds-Transcriber is backed by the awesome cpal for streaming and nnnoiseless for audio cleanup.

Setting Up

You need to obtain the deepspeech-model (tested with 0.9.x) and the native-client for your system and add that folder to your LD_LIBRARY_PATH and LIBRARY_PATH environment variables. See the quick start guide over at deepspeech-rs.

Usage

Add the crate to your Cargo.toml

[dependencies]
ds-transcriber = "0.1.3"

Instantiate your model:

    // the path where your model and native-client lie
    let model_dir_str = args().nth(1).expect("Please specify model dir");
    let mut ds_model = DeepSpeechModel::instantiate_from(model_dir_str);
    let model = ds_model.model();

Create a mutable configuration with your model

    let mut config = ds_transcriber::transcriber::StreamSettings {
        //value used for pause detection, a pause is detected when the amplitude is less than this
        silence_level: 200,
        // takes a reference of the model we instantiated earlier
        model,
        // show the amplitude values on stdout (helps you to find your silence level)
        show_amplitudes: true,
        // seconds of silence indicating end of speech (begin transcription when pause_length is greater than....)
        pause_length: 2.0,
    };

After getting config ready, all you need to do is pass it to transcribe:

    let i_said = ds_transcriber::transcriber::transcribe(&mut config).unwrap();
    println!("I said: {}", i_said);

Repeat the last step to get another transcription with the same configuration.

Contributions

Heck yeah! Pull requests are the greatest thing since sliced bread.

License

MIT

Free Software, Gotta love it!