cognitive-services-speech-sdk-go

Overview

This project contains Golang binding for the Microsoft Cognitive Service Speech SDK.

Getting Started

Check the Speech SDK Setup documentation for Go

Get started with speech-to-text sample for Go

Get started with text-to-speech sample for Go

This project requires Go 1.13

Embedded (offline) speech

The SDK also supports embedded (offline) speech recognition, synthesis and translation, which run fully on-device without using the cloud Speech service. Use speech.NewEmbeddedSpeechConfigFromPath (or NewEmbeddedSpeechConfigFromPaths) to point at the directory that contains your offline models and voices, then reuse the standard recognizer, synthesizer and translation recognizer factories:

config, err := speech.NewEmbeddedSpeechConfigFromPath("/path/to/models")
if err != nil {
    // handle error
}
defer config.Close()

// Select a recognition model and its license text.
config.SetSpeechRecognitionModel("en-US model name", os.Getenv("EMBEDDED_SPEECH_MODEL_LICENSE"))

// The embedded config wraps a regular SpeechConfig, so pass GetSpeechConfig() to the existing factory.
recognizer, err := speech.NewSpeechRecognizerFromConfig(config.GetSpeechConfig(), audioConfig)

// For embedded translation, select a translation model and use the dedicated factory.
config.SetSpeechTranslationModel("translation model name", os.Getenv("EMBEDDED_SPEECH_MODEL_LICENSE"))
translationRecognizer, err := speech.NewTranslationRecognizerFromEmbeddedConfig(config, audioConfig)

Embedded speech has additional runtime requirements:

• The embedded native runtime extensions (for example libMicrosoft.CognitiveServices.Speech.extension.embedded.sr/.tts and their onnxruntime dependency) must be present next to the core Speech SDK library on the load path.
• Licensed speech recognition models, synthesis voices and/or translation models must be installed on the device and passed to the config via the model/voice path.
• Embedded speech is a Limited Access feature that requires approval from Microsoft.

Discovering installed models

If you don't know the exact model or voice name, enumerate what is installed under the configured path and use one of the returned names:

models, err := config.GetSpeechRecognitionModels() // or GetSpeechTranslationModels()
if err != nil {
    // handle error
}
for _, model := range models {
    fmt.Printf("%s (locales: %v)\n", model.Name(), model.Locales())
    model.Close()
}
config.SetSpeechRecognitionModel(models[0].Name(), os.Getenv("EMBEDDED_SPEECH_MODEL_LICENSE"))

Building and running

Embedded speech uses cgo, so the Go toolchain must be told where the native Speech SDK headers and libraries live at build time, and the shared libraries must be discoverable at run time. On Linux/x64, point CGO_CFLAGS/CGO_LDFLAGS at the embedded Speech SDK package and add its lib folder to the loader path:

export SPEECHSDK_ROOT=/path/to/SpeechSDK-Embedded-Linux
export CGO_CFLAGS="-I$SPEECHSDK_ROOT/include/c_api"
export CGO_LDFLAGS="-L$SPEECHSDK_ROOT/lib/x64 -lMicrosoft.CognitiveServices.Speech.core"
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$SPEECHSDK_ROOT/lib/x64"

go build ./...

The samples/embedded package provides runnable functions. Using the sample runner in samples, the arguments are (modelPath, modelOrVoiceName, wavFile) followed by the sample name:

# Speech-to-text
go run . /path/to/models "en-US model name" input.wav embedded:RecognizeOnceFromWavFile

# Speech translation
go run . /path/to/translation/models "translation model name" input.wav embedded:TranslateOnceFromWavFile

# Text-to-speech
go run . /path/to/voices "en-US voice name" output.wav embedded:SynthesisToWavFile

The model/voice license text is read from the EMBEDDED_SPEECH_MODEL_LICENSE environment variable (the same license applies to recognition, synthesis and translation) so it is not passed on the command line.

See the embedded samples guide for prerequisites, native library setup, model/voice installation, build flags and a walkthrough of each sample.

Reference

Reference documentation for these packages is available at http://aka.ms/csspeech/goref

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.