Spaces:
Running
Running
| # imports | |
| import os | |
| os.system("pip install git+https://github.com/openai/whisper.git") | |
| import gradio as gr | |
| import whisper | |
| # the model we are using for ASR, options are small, medium, large and largev2 (large and largev2 don't fit on huggingface cpu) | |
| model = whisper.load_model("small") | |
| # A table to look up all the languages | |
| language_id_lookup = { | |
| "English" : "en", | |
| "German" : "de", | |
| "Greek" : "el", | |
| "Spanish" : "es", | |
| "Finnish" : "fi", | |
| "Russian" : "ru", | |
| "Hungarian" : "hu", | |
| "Dutch" : "nl", | |
| "French" : "fr", | |
| 'Polish' : "pl", | |
| 'Portuguese': "pt", | |
| 'Italian' : "it", | |
| } | |
| # The predict function. audio, language and mic_audio are all parameters directly passed by gradio | |
| # which means they are user inputted. They are specified in gr.inputs[] block at the bottom. The | |
| # gr.outputs[] block will specify the output type. | |
| def predict(audio, language, mic_audio=None): | |
| # checks if mic_audio is used, otherwise feeds model uploaded audio | |
| if mic_audio is not None: | |
| input_audio = mic_audio | |
| elif audio is not None: | |
| input_audio = audio | |
| else: | |
| return "(please provide audio)" | |
| audio = whisper.load_audio(input_audio) | |
| audio = whisper.pad_or_trim(audio) | |
| mel = whisper.log_mel_spectrogram(audio).to(model.device) | |
| if(language == "Detect Language"): | |
| outLanguage, probs = model._detect_language(mel) | |
| print("Detected language is: " + outLanguage) | |
| else: | |
| outLanguage = language_id_lookup[language.split()[0]] | |
| options = whisper.DecodingOptions(fp16 = False, language = outLanguage) | |
| result = whisper.decode(model, mel, options) | |
| print(result.text + " " + outLanguage) | |
| return result.text, outLanguage | |
| title = "Demo for Whisper -> Something -> XLS-R" | |
| description = """ | |
| <b>How to use:</b> Upload an audio file or record using the microphone. The audio is converted to mono and resampled to 16 kHz before | |
| being passed into the model. The output is the text transcription of the audio. | |
| """ | |
| gr.Interface( | |
| fn=predict, | |
| inputs=[ | |
| gr.Audio(label="Upload Speech", source="upload", type="filepath"), | |
| gr.inputs.Dropdown(['English Text', | |
| 'German Text', | |
| 'Greek Text', | |
| 'Spanish Text', | |
| 'Finnish Text', | |
| 'Russian Text', | |
| 'Hungarian Text', | |
| 'Dutch Text', | |
| 'French Text', | |
| 'Polish Text', | |
| 'Portuguese Text', | |
| 'Italian Text', | |
| 'Detect Language'], type="value", default='English Text', label="Select the Language of the that you are speaking in."), | |
| gr.Audio(label="Record Speech", source="microphone", type="filepath"), | |
| ], | |
| outputs=[ | |
| gr.Text(label="Transcription"), | |
| ], | |
| title=title, | |
| description=description, | |
| ).launch() |