import io import math from queue import Queue from threading import Thread from typing import Optional import numpy as np # import spaces import gradio as gr import torch from parler_tts import ParlerTTSForConditionalGeneration from pydub import AudioSegment from transformers import AutoTokenizer, AutoFeatureExtractor, set_seed from transformers.generation.streamers import BaseStreamer device = "cuda:0" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu" torch_dtype = torch.float16 if device != "cpu" else torch.float32 repo_id = "parler-tts/parler_tts_mini_v0.1" model = ParlerTTSForConditionalGeneration.from_pretrained( repo_id, torch_dtype=torch_dtype, attn_implementation="eager", ).to(device) padding_side = "left" model.generation_config.cache_implementation = "static" model.forward = torch.compile(model.forward, fullgraph=False, mode="reduce-overhead") MAX_PROMPT_TOKENS = 70 MAX_DESCRIPTION_TOKENS = 70 description_tokenizer = AutoTokenizer.from_pretrained(repo_id) prompt_tokenizer = AutoTokenizer.from_pretrained(repo_id, padding_side=padding_side) feature_extractor = AutoFeatureExtractor.from_pretrained(repo_id) SAMPLE_RATE = feature_extractor.sampling_rate SEED = 42 default_text = "Please surprise me and speak in whatever voice you enjoy." examples = [ [ "Remember - this is only the first iteration of the model! To improve the prosody and naturalness of the speech further, we're scaling up the amount of training data by a factor of five times.", "A male speaker with a low-pitched voice delivering his words at a fast pace in a small, confined space with a very clear audio and an animated tone.", ], [ "'This is the best time of my life, Bartley,' she said happily.", "A female speaker with a slightly low-pitched, quite monotone voice delivers her words at a slightly faster-than-average pace in a confined space with very clear audio.", ], [ "Montrose also, after having experienced still more variety of good and bad fortune, threw down his arms, and retired out of the kingdom.", "A male speaker with a slightly high-pitched voice delivering his words at a slightly slow pace in a small, confined space with a touch of background noise and a quite monotone tone.", ], [ "Montrose also, after having experienced still more variety of good and bad fortune, threw down his arms, and retired out of the kingdom.", "A male speaker with a low-pitched voice delivers his words at a fast pace and an animated tone, in a very spacious environment, accompanied by noticeable background noise.", ], ] sampling_rate = model.audio_encoder.config.sampling_rate frame_rate = model.audio_encoder.config.frame_rate # @spaces.GPU def generate_base(text, description): inputs = description_tokenizer(description, return_tensors="pt", padding='max_length', max_length=MAX_DESCRIPTION_TOKENS).to(device) prompt = prompt_tokenizer(text, return_tensors="pt", padding='max_length', max_length=MAX_PROMPT_TOKENS).to(device) generation_kwargs = dict( input_ids=inputs.input_ids, prompt_input_ids=prompt.input_ids, attention_mask=inputs.attention_mask, prompt_attention_mask=prompt.attention_mask, do_sample=True, temperature=1.0, min_new_tokens=10, ) new_audio = model.generate(**generation_kwargs).detach()[0].float().cpu().numpy() print(new_audio) # new_audio = new_audio / np.abs(new_audio).max() # new_audio = new_audio * 32767 # new_audio = new_audio.astype(np.int16) return sampling_rate, new_audio css = """ #share-btn-container { display: flex; padding-left: 0.5rem !important; padding-right: 0.5rem !important; background-color: #000000; justify-content: center; align-items: center; border-radius: 9999px !important; width: 13rem; margin-top: 10px; margin-left: auto; flex: unset !important; } #share-btn { all: initial; color: #ffffff; font-weight: 600; cursor: pointer; font-family: 'IBM Plex Sans', sans-serif; margin-left: 0.5rem !important; padding-top: 0.25rem !important; padding-bottom: 0.25rem !important; right:0; } #share-btn * { all: unset !important; } #share-btn-container div:nth-child(-n+2){ width: auto !important; min-height: 0px !important; } #share-btn-container .wrap { display: none !important; } """ with gr.Blocks(css=css) as block: gr.HTML( """

Parler-TTS 🗣️

""" ) gr.HTML( f"""

Parler-TTS is a training and inference library for high-fidelity text-to-speech (TTS) models. Two models are demonstrated here, Parler-TTS Mini v0.1, is the first iteration model trained using 10k hours of narrated audiobooks, and Parler-TTS Jenny, a model fine-tuned on the Jenny dataset. Both models generates high-quality speech with features that can be controlled using a simple text prompt (e.g. gender, background noise, speaking rate, pitch and reverberation).

Tips for ensuring good generation:

""" ) with gr.Tab("Base"): with gr.Row(): with gr.Column(): input_text = gr.Textbox(label="Input Text", lines=2, value=default_text, elem_id="input_text") description = gr.Textbox(label="Description", lines=2, value="", elem_id="input_description") run_button = gr.Button("Generate Audio", variant="primary") with gr.Column(): audio_out = gr.Audio(label="Parler-TTS generation", elem_id="audio_out", autoplay=True) inputs = [input_text, description] outputs = [audio_out] gr.Examples(examples=examples, fn=generate_base, inputs=inputs, outputs=outputs, cache_examples=True) run_button.click(fn=generate_base, inputs=inputs, outputs=outputs, queue=True) gr.HTML( """

To improve the prosody and naturalness of the speech further, we're scaling up the amount of training data to 50k hours of speech. The v1 release of the model will be trained on this data, as well as inference optimisations, such as flash attention and torch compile, that will improve the latency by 2-4x. If you want to find out more about how this model was trained and even fine-tune it yourself, check-out the Parler-TTS repository on GitHub. The Parler-TTS codebase and its associated checkpoints are licensed under Apache 2.0.

""" ) block.queue() block.launch(share=True)