edgetts

edgetts is a golang module that allows you to use Microsoft Edge's online text-to-speech service in your golang projects.

Installation

To install it, run the following command:

go get github.com/kolonist/edgetts@latest

Usage

Speech synthesys to file

package main

import (
	"context"
	"github.com/kolonist/edgetts"
)

func main() {
	args := edgetts.Args{
		// set voice to use in speech synthesys
		Voice: "en-US-AlloyTurboMultilingualNeural",
	}

	// generate sound in mp3 format and save it to the file
	err := edgetts.
		New(args).
		Speak("Text I need to speak now").
		SaveToFile(context.TODO(), "./sample.mp3", edgetts.OutputFormatMp3)
}

Speech metadata

package main

import (
	"context"
	"github.com/kolonist/edgetts"
)

func main() {
	args := edgetts.Args{
		// set voice to use in speech synthesys
		Voice: "en-US-AlloyTurboMultilingualNeural",

		// speak 15% faster
		Rate:  "+15%",

		// reduce the volume by 10%
		Volume: "-10%",
	}

	// set synthesys args and text
	speaker := edgetts.
		New(args).
		Speak("Text I need to speak now")

	// generate sound in raw PCM format with 22050 Hz sample rate and get sound chunks in iterator
	for data, err := range speaker.GetSoundIter(context.TODO(), edgetts.OutputFormatRaw22050) {
		// data is `[]byte`
		// you can use this method to stream audio as it came from Edge TTS server
	}

	// get info about timings of all words appearance and its duration in generated speech
	// note that you should call `GetMetadata()` after one of `GetSoundIter()`, `GetSound()` or `SaveToFile()`
	metadata, err := speaker.GetMetadata()

	for _, word := range metadata {
		// `Offset` and `Duration` are in milleseconds
		fmt.Printf("%d: %s - %d\n", word.Offset, word.Text, word.Duration)
	}
}

Getting list of voices

package main

import (
	"context"
	"github.com/kolonist/edgetts"
)

func main() {
	// get ;list of all available voices
	voices, err := edgetts.ListVoices(context.TODO())

	for i, v := range voices {
		// you should use `ShortName` in your speech generation Args
		fmt.Printf("ShortName: %s, Gender: %s", v.ShortName, v.Gender)
	}
}

You can find more complex example in /examples folder.

API

Constants:

Audio output formats (OutputFormat):

• OutputFormatMp3 — mp3 24khz, 48k bitrate (default)
• OutputFormatWebm — webm 24khz, 16bit, 24k bitrate
• OutputFormatOgg — ogg 24khz, 16bit
• OutputFormatRaw22050 — raw PCM 22050 hz, 16bit
• OutputFormatRaw44100 — raw PCM 44100 hz, 16bit

Structs:

edgetts.Args

Arguments to use in edgetts.New(args edgetts.Args) function.

Fields:

• Voice string — Voice.
  Has format en-US-AlloyTurboMultilingualNeural, where:
  • en-US — locale
  • AlloyTurbo - actual voice name
• Volume string — Sound volume in percent. Can increase (+10%) or decrease (-20%) volume
• Rate string — Speech rate in percent. Can increase (+30%) or decrease (-40%) rate

edgetts.EdgeTTS

Struct to manage speech synthesys

Constructor

New(args edgetts.Args) *edgetts.EdgeTTS

Create EdgeTTS struct

Methods:

Speak(text string) *Speaker

Assign text you need to synthesize.

SpeakWithVoice(text string, voice string) *Speaker

Assign text you need to synthesize with defined voice. Can be helpful if you need multiple generations with different voices.

edgetts.Speaker

Used to synthesyze speech

Methods:

GetSoundIter(ctx context.Context, format OutputFormat) iter.Seq2[[]byte, error]

Get sound data as byte buffers in iterator

GetSound(ctx context.Context, format OutputFormat) ([]byte, error)

Get whole downloaded sound file as byte buffer

SaveToFile(ctx context.Context, filename string, format OutputFormat) error

Save to file generated sound

GetMetadata() ([]SpeechMetadata, error)

Get metadata of generated speech. Should be called after one of GetSoundIter(), GetSound() or SaveToFile()

edgetts.SpeechMetadata

Contains time of each word start and its pronunciation duration im milliseconds

Fields:

• Offset int — Start time of word in generated sound in milliseconds
• Duration int — Duration of word pronunciation in milliseconds
• Text string — Word

edgetts.Voice

Voice used for speech synthesys

Get list of voices:

edgetts.ListVoices(ctx context.Context) ([]Voice, error)

Get list of all possible voices

Fields:

• Name string — Voice full name
• ShortName string — Voice short name. You should use this value when specifying voice in this library functions
• Gender string — Speaker gender, Male of Female
• Locale string — Locale, e.g. en-US
• SuggestedCodec string — always empty
• FriendlyName string — always empty
• Status string — Can be GA for General Availability or Preview
• VoiceTag.ContentCategories []string — always empty
• VoiceTag.VoicePersonalities []string — Vocal characteristics of voice

For speech synthesys you need only ShortName field

Thanks

I used the following projects as sources of inspiration:

• https://github.com/rany2/edge-tts (similar library for Python)
• https://github.com/surfaceyu/edge-tts-go (Python library rewritten in Go but not working now)

Author

@license MIT
@author Alexander Zubakov developer@xinit.ru