text-to-speech is not a valid pipeline

#7
by boddles2 - opened

When I run the "Hosted inference API" I get text-to-speech is not a valid pipeline

cc @Matthijs is SpeechT5 supported by a pipeline?

Indeed, there currently is no pipeline for the TTS task in Transformers.

:disappointed_face:

Update, a TTS pipeline will be added once at least 2 different TTS models are present in the Transformers library. This is to ensure the design is robust enough to handle different models

Update, a TTS pipeline will be added once at least 2 different TTS models are present in the Transformers library. This is to ensure the design is robust enough to handle different models

What is required to accomplish this?

Bump, problem still persists.

A text-to-audio pipeline is now available: https://github.com/huggingface/transformers/pull/24952, supporting SpeechT5 and Bark. Usage is as follows:

from transformers import pipeline

classifier = pipeline(model="suno/bark")
output = pipeline("Hey it's HuggingFace on the phone!")
audio = output["audio"]
sampling_rate = output["sampling_rate"]

Next step is to create a corresponding inference widget for it, cc @mishig .

Sign up or log in to comment