Speech Recognition Interface

Setting up the speech recognition application to detect a wakeup word may be done using series of Audio Elements linked into a pipeline shown below.

Sample Speech Recognition Pipeline

Configuration and use of particular elements is demonstrated in several examples linked to elsewhere in this documentation. What may need clarification is use of the Filter and the RAW stream. The filter is used to adjust the sample rate of the I2S stream to match the sample rate of the speech recognition model. The RAW stream is the way to feed the audio input to the model.

Application Example

The speech_recognition/asr/main/example_asr_main.c example demonstrates how to initialize the model, determine the number of samples and the sample rate of voice data to feed to the model, and detect the wakeup word.

Implementation of the speech recognition API is demonstrated in that example.

API Reference

For the latest API reference please refer to Espressif Speech recognition repository.