Spaces:
Paused
Paused
| # import spaces | |
| import tempfile | |
| import wave | |
| import gradio as gr | |
| import os | |
| import re | |
| import torch | |
| import soundfile as sf | |
| import numpy as np | |
| import torch.nn.functional as F | |
| from whisperspeech.pipeline import Pipeline | |
| from whisperspeech.languages import LANGUAGES | |
| from whisperspeech.pipeline import Pipeline | |
| from whisperspeech.utils import resampler | |
| title = """# 🙋🏻♂️ Welcome to🌟Collabora🌬️💬📝WhisperSpeech | |
| You can use this ZeroGPU Space to test out the current model [🌬️💬📝collabora/whisperspeech](https://huggingface.co/collabora/whisperspeech). 🌬️💬📝collabora/whisperspeech is An Open Source text-to-speech system built by inverting Whisper. Install it and use your command line interface locally with `pip install whisperspeech`. It's like Stable Diffusion but for speech – both powerful and easily customizable : so you can use it programmatically in your own pipelines! [Contribute to whisperspeech here](https://github.com/collabora/WhisperSpeech) | |
| You can also use 🌬️💬📝WhisperSpeech by cloning this space. 🧬🔬🔍 Simply click here: <a style="display:inline-block" href="https://huggingface.co/spaces/Tonic/laion-whisper?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a></h3> | |
| We're **celebrating the release of the whisperspeech** at [the LAION community, if you love open source ai learn more here : https://laion.ai/](https://laion.ai/) big thanks to the folks at huggingface for the community grant 🤗 | |
| ### How to Use | |
| Input text with tahe language identifiers provided to create a multilingual speech. Optionally you can add an audiosample to make a voice print.Scroll down and try the api <3 Gradio. | |
| This space runs on ZeroGPU, so **you need to be patient** while you acquire the GPU and load the model the first time you make a request ! | |
| """ | |
| # text examples=["<en> Hello, how are you? <fr> Bonjour, comment ça va?", "<de> Guten Tag <it> Buongiorno <jp> こんにちは"] | |
| # audio examples=["path/to/tonic.wav"] | |
| # Function to parse the multilingual input text | |
| def parse_multilingual_text(input_text): | |
| pattern = r"<(\w+)>\s(.*?)\s(?=<\w+>|$)" | |
| segments = re.findall(pattern, input_text) | |
| return [(lang, text.strip()) for lang, text in segments if lang in LANGUAGES.keys()] | |
| #@spaces.GPU | |
| def generate_segment_audio(text, lang, speaker_url, pipe): | |
| if not isinstance(text, str): | |
| text = text.decode("utf-8") if isinstance(text, bytes) else str(text) | |
| # Generating stoks (tokens<pl>) from text | |
| # stoks = pipe.t2s.generate([text], lang=[lang]) | |
| audio_data = pipe.generate(text, speaker_url, lang) | |
| resample_audio = resampler(newsr=24000) | |
| audio_data_resampled = next(resample_audio([{'sample_rate': 24000, 'samples': audio_data.cpu()}]))['samples_24k'] | |
| audio_np = audio_data_resampled.cpu().numpy() | |
| print("Shape after resampling:", audio_np.shape) # Debug statement | |
| return audio_np | |
| # Function to append and concatenate audio segments with padding | |
| def concatenate_audio_segments(segments): | |
| # # Determine the length of the longest segment | |
| # max_length = max(seg.shape[0] for seg in segments) | |
| # print("Max length of segments:", max_length) # Debug statement | |
| # # Pad each segment to the length of the longest segment and stack them | |
| # padded_segments = [] | |
| # for seg in segments: | |
| # # Check if the segment is stereo; if not, convert it to stereo | |
| # if seg.ndim == 1 or seg.shape[1] == 1: | |
| # stereo_segment = np.stack((seg, seg), axis=-1) | |
| # else: | |
| # stereo_segment = seg | |
| # Pad the segment to the max length | |
| # padding_length = max_length - stereo_segment.shape[0] | |
| # padded_segment = np.pad(stereo_segment, ((0, padding_length), (0, 0)), 'constant') | |
| # print("Padded segment shape:", padded_segment.shape) # Debug statement | |
| # padded_segments.append(padded_segment) | |
| concatenated_audio = np.concatenate(segments , axis=1) | |
| print("Concatenated audio shape:", concatenated_audio.shape) # Debug statement | |
| # concatenated_audio = concatenated_audio / np.max(np.abs(concatenated_audio)) | |
| return concatenated_audio | |
| # The rest of the code in app.py remains the same | |
| # @spaces.GPU | |
| def whisper_speech_demo(multilingual_text, speaker_audio): | |
| segments = parse_multilingual_text(multilingual_text) | |
| if not segments: | |
| return None, "No valid language segments found. Please use the format: <lang> text" | |
| pipe = Pipeline() | |
| speaker_url = speaker_audio if speaker_audio is not None else None | |
| audio_segments = [] | |
| for lang, text in segments: | |
| text_str = text if isinstance(text, str) else str(text) | |
| audio_np = generate_segment_audio(text_str, lang, speaker_url, pipe) | |
| print("Audio segment shape:", audio_np.shape) # Debug statement | |
| audio_segments.append(audio_np) | |
| # Normalize the concatenated audio | |
| concatenated_audio = concatenated_audio / np.max(np.abs(concatenated_audio)) | |
| # Write the audio data to a temporary file and return the file path | |
| with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file: | |
| sf.write(tmp_file.name, concatenated_audio.T, 24000, format='WAV', subtype='PCM_16') | |
| return tmp_file.name | |
| with gr.Blocks() as demo: | |
| gr.Markdown(title) | |
| output_audio = gr.Audio(label="Generated Speech") | |
| generate_button = gr.Button("Try 🌟Collabora🌬️💬📝WhisperSpeech") | |
| with gr.Row(): | |
| text_input = gr.Textbox(label="Enter multilingual text", placeholder="e.g., <en> Hello <fr> Bonjour <es> Hola") | |
| speaker_input = gr.Audio(label="Upload or Record Speaker Audio (optional)", sources=["upload", "microphone"]) | |
| with gr.Accordion("Available Languages and Their Tags"): | |
| language_list = "\n".join([f"{lang}: {LANGUAGES[lang]}" for lang in LANGUAGES]) | |
| gr.Markdown(language_list) | |
| generate_button.click(whisper_speech_demo, inputs=[text_input, speaker_input], outputs=output_audio) | |
| demo.launch() |