According to https://aws.amazon.com/what-is/speech-to-text:

Speech to text is a speech recognition software that enables the recognition and translation of spoken language into text through computational linguistics. It is also known as speech recognition or computer speech recognition. Specific applications, tools, and devices can transcribe audio streams in real-time to display text and act on it.

Open Voice OS provides support for different STT engines via a plugin mechanism exposing HTTP endpoints to be consumed by the voice assistant.

To facilitate the installation and the adoption of local Speech-to-Text engine, we build a set of OCI images compatible with Docker, Podman and Kubernetes as well.

Using this approach allows you as well to decentralize the STT server which means that it doesn't have to run locally on the voice assistant but on a remote server with more compute power using CPU and/or GPU.

There are two (2) different implementations for the Faster Whisper STT plugin.

• ovos-stt-plugin-fasterwhisper image using only the CPU to transcribe (default)
• ovos-stt-plugin-fasterwhisper-cuda image using only the GPU to transcribe

To use ovos-stt-plugin-fasterwhisper-cuda, please review the docker-compose.yml file.

Only one implementation can be selected at a time.

Docker or Podman (rootless) is of course required and docker compose (not docker-compose!!) or podman-compose is a nice to have to simplify the whole process of deploying the whole stack by using the docker-compose.yml files (for Docker, this command will be embedded depending the version, for Podman, podman-compose command comes from a different package).

If you plan to passthrough GPUs in order to leverage Nvidia CUDA with Docker or Podman, please make you configured your container engine properly to support GPUs.

The base image is the main layer for the other images, for example the fasterwhisper image requires the base image to be build.

git clone https://github.com/OpenVoiceOS/ovos-docker-stt.git
cd ovos-docker-stt
docker buildx build fasterwhisper/ -t smartgic/ovos-stt-server-fasterwhisper:alpha --build-arg BUILD_DATE=$(date -u +'%Y-%m-%dT%H:%M:%SZ') --no-cache
  # Or:
podman buildx build fasterwhisper/ -t smartgic/ovos-stt-server-fasterwhisper:alpha --build-arg BUILD_DATE=$(date -u +'%Y-%m-%dT%H:%M:%SZ') --no-cache

There are a list of available arguments that could be used during the image build process.

Pre-build images are already available here and are the default referenced within the docker-compose.yml file.

docker-compose.yml file provides an easy way to provision the container stack (volumes and services) with the required configuration for each of them. docker compose or podman-compose both support environment files, check the .env file.

git clone https://github.com/OpenVoiceOS/ovos-docker-stt.git
mkdir -p ~/ovos-tts-stt/config
chown ${USER}:${USER} -R ~/ovos-tts-stt
cd ovos-docker-stt
docker compose up -d
  # Or:
podman-compose up -d

To reduce the potential overhead due to the image downloads and extracts, the --parallel option could be user in order to process the images by batch of x (where x is an integer).

docker compose --parallel 3 up -d
  # Or:
podman-compose --parallel 3 up -d

If you only plan to use the Faster Whisper STT server then you could reference it to the command line.

docker compose up -d ovos_stt_fasterwhisper
  # Or:
podman-compose up -d ovos_stt_fasterwhisper

Some variables might need to be tuned to match your setup such as the timezone, the directories, etc..., have a look into the .env files befor running docker compose or podman-compose.

The OVOS_USER variable should be changed only if you build the Docker images with a different user than ovos.

The easiest way to update a stack already deployed by docker compose or podman-compose is to use docker compose or podman-compose. ☺️

Because the pull_policy option of each service is set to always, everytime that a new image is uploaded with the same tag then docker compose or podman-compose will pull-it and re-create the container based on this new image.

docker compose up -d
  # Or:
podman-compose up -d

If you want to change the tag to deploy, update the .env file with the new value.

~/ovos/config/mycroft.conf configuration file is used to configura the STT plugin. Make sure to adapt the sample below to fit your requirements.

{
  "logs": {
    "path": "stdout"
  },
  "stt": {
    "module": "ovos-stt-plugin-fasterwhisper",
    "ovos-stt-plugin-fasterwhisper": {
        "model": "whisper-large-v3-turbo",
        "compute_type": "float16",
        "use_cuda": true,
        "cpu_thread": 8
    },
    "ovos-stt-plugin-vosk-streaming": {
        "model": "https://alphacephei.com/vosk/models/vosk-model-en-us-0.42-gigaspeech.zip",
        "verbose": false
    },
    "ovos-stt-plugin-vosk": {
        "model": "http://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip",
        "verbose": false
    },
    "ovos-stt-plugin-deepgram": {
      "key": "GET A KEY FROM DEEPGRAM WEBSITE :)"
    },
    "ovos-stt-plugin-chromium": {
        "lang": "en-US",
        "pfilter": false,
        "debug": false
    }
  }
}

If you don't plan to use Nvidia CUDA with the STT Faster Whisper plugin, then use_cuda should be set to false and compute_type set to int8.

Once the STT servers are up and running, the voice assistant must be configured to reference them. Please make sure to add the section below to your ~/ovos/config/mycroft.conf configuration file.

{
  "stt": {
      "module": "ovos-stt-plugin-server",
      "fallback_module": "ovos-stt-plugin-vosk",
      "ovos-stt-plugin-server": {
        "urls": [
          "http://192.168.1.227:8080/stt",
          "http://192.168.1.227:8081/stt",
          "http://192.168.1.227:8082/stt",
          "http://192.168.1.227:8083/stt",
          "http://192.168.1.227:8084/stt",
          "https://stt.openvoiceos.org/stt"
        ]
      }
  }
}

The configuration means that ovos-stt-plugin-server will be used as default STT plugin. The plugin has a list of five (5) STT servers, if one is down then the plugin goes to the next one, etc...

If all the STT servers from ovos-stt-plugin-server are down then the voice assistant will fallback to the ovos-stt-plugin-vosk STT server running locally to the voice assistant.

In order to check if a STT server is up and running, the /status endpoint should be called (jq command is not mandatory just nice to have).

curl -v http://192.168.1.227:8080/status | jq

Enable debug mode in ~/ovos/config/mycroft.conf to get more verbosity from the logs. All containers will have to be restarted to receive the configuration change.

{
  "debug": true,
  "log_level": "DEBUG",
  "logs": {
    "path": "stdout"
  }
}

To access all the container logs at the same time, run the following command (make sure it matches the docker compose or podman-compose command you run to deploy the stack):

docker compose logs -f --tail 200
  # Or:
podman-compose logs -n -f --tail 200

To access the logs of a specific container, run the following command:

docker logs -f --tail 200 ovos_stt_fasterwhisper
  # Or:
podman logs -f --tail 200 ovos_stt_fasterwhisper

To go inside a container and run multiple commands, run the following command (where bash is the available shell in there):

docker exec -ti ovos_stt_fasterwhisper bash
  # Or:
podman exec -ti ovos_stt_fasterwhisper bash

If the configuration file is not valid JSON, jq will return something like this:

parse error: Expected another key-value pair at line 81, column 3

To get the CPU, memory and I/O consumption per container, run the following command:

docker stats -a --no-trunc
  # Or:
podman stats -a --no-trunc

Make sure mycroft.conf configuration file is JSON valid by using the jq command.

cat ~/ovos/config/mycroft.conf | jq

• Matrix channel
• Open Voice OS documentation
• Contribute to Open Voice OS
• Report bugs related to these Docker images