Persian-IPA-to-Speech-F5

A small Python API for the KiaBush/Persian-IPA-to-Speech-F5 checkpoint. It downloads the model on first use, preserves raw Unicode IPA, and hides the compatibility details of the upstream F5-TTS API.

Install

pip install persian-ipa-to-speech-f5

FFmpeg must be available on the system. For GPU inference, install the PyTorch build matching your CUDA version before installing this package.

IPA to speech

from persian_ipa_to_speech_f5 import PersianIPAToSpeechF5

tts = PersianIPAToSpeechF5(device="cuda")
tts.synthesize(
    "sælɒːm, xoːbiː?",
    reference_audio="reference.wav",
    reference_ipa="<exact IPA transcription of the reference audio>",
    output="outputs/result.wav",
)

The reference IPA must accurately transcribe the spoken reference audio and follow the checkpoint's IPA convention.

Persian text to speech

The optional front end uses KiaBush/persian-text-to-ipa-byt5:

from persian_ipa_to_speech_f5 import PersianIPAToSpeechF5, PersianTextToIPA

ipa = PersianTextToIPA(device="cpu").convert("گل رز واقعاً زیباست.")

tts = PersianIPAToSpeechF5(device="cuda")
tts.synthesize(
    ipa,
    reference_audio="reference.wav",
    reference_ipa="<exact IPA transcription of the reference audio>",
    output="outputs/result.wav",
)

Names, numbers, abbreviations, and dialectal input can be context-dependent; review generated IPA when pronunciation matters.

Command line

persian-ipa-to-speech-f5 \
  --text "گل رز واقعاً زیباست." \
  --reference-audio reference.wav \
  --reference-ipa "<reference IPA>" \
  --output outputs/result.wav \
  --device cuda

Use --ipa instead of --text to bypass the text-to-IPA model.

Licenses

The source code in this Python package is licensed under Apache-2.0. The model checkpoint is a derivative of F5-TTS v1 Base and is separately licensed under CC-BY-NC-4.0; its non-commercial restriction applies to the model itself. Rights in generated audio can depend on the input and applicable law. F5-TTS source code is MIT-licensed. Users are responsible for obtaining the rights and consent needed for reference voices and audio.

Authors and contributors

• Ali Akbar Kiaei — research, model training, engineering, and release
• Mahnaz Bush — data preparation, model training, evaluation, research, and documentation