str2speech

👉 str2speech

Overview

str2speech is a simple command-line tool for converting text to speech using Transformer-based text-to-speech (TTS) models. It supports multiple models and voice presets, allowing users to generate high-quality speech audio from text.

Latest

We just added support for Dia-1.6B and Hindi voices in Kokoro TTS.

We just added support for ByteDance's MegaTTS3. Here's how easy it is to use it:

str2speech --model megatts3 --text "This is awesome!"

Works fine with just a CPU (needs about 10GB of RAM). But it's always better if you have CUDA available.

We now support Microsoft's Speech T5. This is a very lightweight model, and sounds pretty good. Try it out with this:

str2speech --model "microsoft/speecht5_tts" \
    --text "My dog is prettier than yours." \
    --output "t5test.wav"

We now support Spark-TTS-0.5B. This is an awesome model. Here's how you use it:

str2speech --model "SparkAudio/Spark-TTS-0.5B" \
        --text "Hello from Spark" \
        --output "sparktest.wav"

Added support for Sesame CSM-1B. Here's how to use it:

export HF_TOKEN=<your huggingface token>
str2speech --text "Hello from Sesame" --model "sesame/csm-1b"

Added support for Kokoro-82M. This is how you run it:

str2speech --text "Hello again" --model "kokoro"

This is probably the easiest way to use Kokoro TTS.

Added support for Zyphra Zonos. Try this out:

str2speech --text "Hello from Zonos" \
    --model "Zyphra/Zonos-v0.1-transformer" \
    --output hellozonos.wav

Alternatively, you could write Python code to use it:

from str2speech.speaker import Speaker

speaker = Speaker("Zyphra/Zonos-v0.1-transformer")
speaker.text_to_speech("Hello, this is a test!", "output.wav")

You might need to install espeak. Here's how you can install it:

sudo apt install espeak-ng

Features

• Supports multiple TTS models, including Sesame/CSM-1B, SparkAudio/Spark-TTS-0.5B, Kokoro, and various facebook/mms-tts models.
• Supports voice cloning with Spark-TTS and Zyphra Zonos.
• Allows selection of voice presets.
• Supports text input via command-line arguments or files.
• Outputs speech in .wav format.
• Works with both CPU and GPU.

Available Models

The following models are supported:

• Sesame/CSM-1B
• MegaTTS3
• SparkAudio/Spark-TTS-0.5B
• Zyphra/Zonos-v0.1-transformer
• microsoft/speecht5_tts
• Kokoro (English, Hindi, and Spanish only)
• suno/bark-small (default TTS model)
• suno/bark
• facebook/mms-tts-eng (English only)
• facebook/mms-tts-deu (German only)
• facebook/mms-tts-fra (French only)
• facebook/mms-tts-spa (Spanish only)
• facebook/mms-tts-swe (Swedish only)
• nari-labs/dia-1.6b

Installation

To install str2speech, first make sure you have pip installed, then run:

pip install str2speech

Usage

Command Line

Run the script via the command line:

str2speech --text "Hello, world!" --output hello.wav

Options

• --text (-t): The text to convert to speech.
• --file (-f): A file containing text to convert to speech.
• --voice (-v): The voice preset to use (optional, defaults to a predefined voice).
• --output (-o): The output .wav file name (optional, defaults to output.wav).
• --model (-m): The TTS model to use (optional, defaults to suno/bark-small).
• --speed (-s): The speed of the speech (optional, defaults to 1.0). Supported only by Kokoro TTS currently.
• --clone (-c): The filename of a wav file that contains the voice to clone.
• --clone-voice-text (-p): The transcript of what's being said in the wav file provided.

Example:

str2speech --file input.txt --output speech.wav --model suno/bark

Example 2:

str2speech --text "This is my cloned voice" \
        --model zyphra/zonos-v0.1-transformer \
        --output clonetest.wav --clone "./lex.wav"

API Usage

You can also use str2speech as a Python module:

from str2speech.speaker import Speaker

speaker = Speaker()
speaker.text_to_speech("Hello, this is a test.", "test.wav")

Tested With These Dependencies

• transformers==4.49.0
• torch==2.5.1+cu124
• numpy==1.26.4
• scipy==1.13.1

License

This project is licensed under the GNU General Public License v3 (GPLv3).