Rhasspy ASR DeepSpeech Hermes

MQTT service for Rhasspy that uses Mozilla's DeepSpeech 0.9.3.

Requirements

• Python 3.7
• Mozilla DeepSpeech 0.9.3
• Pre-trained model
• generate_scorer_package in PATH from native client
• build_binary in PATH from KenLM

Installing

Clone the repository and create a virtual environment:

$ git clone https://github.com/rhasspy/rhasspy-asr-deepspeech-hermes.git
$ cd rhasspy-asr-deepspeech-hermes
$ ./configure
$ make
$ make install

Running

Run script:

bin/rhasspy-asr-deepspeech-hermes \
    --model /path/to/output_graph.pbmm \
    --language-model /path/to/lm.binary \
    --trie /path/to/trie \
    --host <MQTT_HOST> \
    --port <MQTT_PORT> \
    --debug

Using

Set Rhasspy ASR system to "Hermes MQTT". Connect Rhasspy and DeepSpeech service to the same MQTT broker (use port 12183 for Rhasspy's internal broker).