Silero-TTS is the simplest solution and supports different models for different languages like English, Spanish, French, German and others. Silero-TTS supports SSML by which you can change the behaviour in the text.
Supported Tags are:
- break
<break time="2000ms" strength="x-weak"/>where time can be in milliseconds (ms) or seconds (s) and strength can be x-weak, weak, medium, strong, x-strong - prosody
<prosody rate="x-slow" pitch="x-high">where rate can be x-slow, slow, medium, fast, x-fast, and pitch can be x-low, low, medium, high, x-high - p
<p>text</p>Represents a paragraph, equivalent to x-strong pause. - s
<s>text</s>Represents a sentence, equivalent to strong pause.
F5-TTS is a TTS Model that supports voice cloning based on an audio sample with fast inference and Multi-Style / Multi-Speaker Generation.
The speed can be set globally in the Settings.
To add your own voice, go to the .cache\f5tts-cache\voices directory
-
Copy a .wav sample audio of the voice with a .txt file with the same name containing the transcript of the spoken text into it.
Best results should be audio files as PCM S16 LE, Mono with a sample rate of 24000 Hz and 16 Bits per sample.
-
If an audio file does not give good results, make sure the transcript is good. Sometimes it also helps to cut the audio shorter.
To generate audio with different speakers, you can add the Speaker name at the beginning of a line like this:
[Justin] This is the text, spoken by the Justin speaker.
[Announcer_Ahri] And this text will be spoken by the Announcer_Ahri voice.