Resemblyzer
speechbrain
Resemblyzer | speechbrain | |
---|---|---|
4 | 26 | |
2,596 | 7,892 | |
1.0% | 2.5% | |
3.4 | 9.8 | |
7 months ago | 7 days ago | |
Python | Python | |
Apache License 2.0 | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Resemblyzer
-
Build an Audio-Driven Speaker Recognition System Using Open-Source Technologies — Resemblyzer and QdrantDB.
Resemblyzer allows us to derive high-level representation of voice through a deep learning model. It simplifies the life of developers by enabling them to convert audio clips into vectors with just a few lines of code, eliminating the need for neural networks. See official github repository.
-
Get timestamps for .wav partials;
I want to perform a modification to Resemblyzer's speaker diarization script to cut out parts of audio where a specific speaker isn't present. While the graph generated by the original demo seems alright, the timestamps at which it chooses to cut the audio are off. I got this conclusion because when I outputted all the timestamp information, my 22 minute long video came out to be 1036 seconds long. Also, the variable I'm indexing the time by seems to be a collection of "wave partials as a list of slices ", as represented by the function that generates its value. Furthermore, the function I was modifying to get the time said that the intervals were non-reliable. This is bad, because as you will see below in my code, when cutting the video with ffmpeg, I treat them as if these were one-to-one with the video: from resemblyzer import preprocess_wav, VoiceEncoder from demo_utils import * from pathlib import Path from os import listdir, system from os.path import join def Diarization(path, file, segments): wav_fpath = Path(join(path, file)) wav = preprocess_wav(wav_fpath) speaker_names = ["Peter"] speaker_wavs = [wav[int(s[0] * sampling_rate):int(s[1]) * sampling_rate] for s in segments] encoder = VoiceEncoder("cpu") print("Running the continuous embedding on cpu, this might take a while...") _, cont_embeds, wav_splits = encoder.embed_utterance(wav, return_partials=True, rate=16) speaker_embeds = [encoder.embed_utterance(speaker_wav) for speaker_wav in speaker_wavs] similarity_dict = {name: cont_embeds @ speaker_embed for name, speaker_embed in zip(speaker_names, speaker_embeds)} times = [((s.start + s.stop) / 2) / sampling_rate for s in wav_splits] keep = True cutTimes = [[times[0], times[len(wav_splits) - 1]]] #similar = open("similarities.txt", "w+") for i in range(len(wav_splits)): similarities = [s[i] for s in similarity_dict.values()] best = np.argmax(similarities) name, similarity = list(similarity_dict.keys())[best], similarities[best] #similar.write(f"{times[i]} - {similarity}\n") if similarity > 0.65: if not keep: cutTimes.append([times[i], times[len(wav_splits) - 1]]) keep = True else: ⠀ if keep: cutTimes[len(cutTimes) - 1][1] = times[i] keep = False #similar.close() cutCommand = "" for num, seg in enumerate(cutTimes): if num == 0: cutCommand += f"between(t,{seg[0]},{seg[1]})" continue cutCommand += f"+between(t,{seg[0]},{seg[1]})" addMe = "Cut - " print(f"ffmpeg -i \"{join(path, file)}\" -af \"aselect='{cutCommand}',asetpts=N/SR/TB\" \"{join(path, addMe+file)}\"") system(f"ffmpeg -y -i \"{join(path, file)}\" -af \"aselect='{cutCommand}',asetpts=N/SR/TB\" \"{join(path, addMe+file)}\"") path = r'C:\Users\mlfre\OneDrive\Desktop\Resemblyzer\Resemblyzer-master\audio_data' for file in listdir(path): #if ".mp3" in file or ".wav" in file or ".mp4" in file: if file == "peter.mp3": segments = [[12, 21]] Diarization(path, file, segments) Since the graph's values were accurate in real time, if I could just manage to get the time intervals accurate in real time as well, I would be golden. Unfortunately, I do not know how to translate from iterating over a list of wav partials as slices, to the length in time of a wave file.
-
[D] state of art for Speaker Diarization?
I've tried Resemblyzer's method, yet it always either cut out too much of his voice, or included too much of others. It also required that i have a clip of him talking, and the quality of that clip heavily impacted its performance.
-
Is there a python based speaker diarization system you would recommend?
Try this: https://github.com/resemble-ai/Resemblyzer
speechbrain
- SpeechBrain 1.0: A free and open-source AI toolkit for all things speech
- FLaNK Stack Weekly 22 January 2024
-
[D] Training ASR model using SpeechBrain
You likely have a very broken sample in one of your batches. It looks like your training actually went through a few batches before it horked the error at you. A quick google shows a similar issue in the github repo: https://github.com/speechbrain/speechbrain/issues/649 .
-
Whisper.cpp
https://github.com/ggerganov/whisper.cpp https://speechbrain.github.io/
-
[D] What is the best open source text to speech model?
I don't know if it's the best, but Speechbrain is supposed to be state of the art.
-
[D] What's stopping you from working on speech and voice?
- https://github.com/speechbrain/speechbrain
- Specific Voice recognition
- How to get high-quality, low-cost Speech-to-Text transcription?
- [D] Speech Enhancement SOTA
- Speaker diarization
What are some alternatives?
pyannote-audio - Neural building blocks for speaker diarization: speech activity detection, speaker change detection, overlapped speech detection, speaker embedding
espnet - End-to-End Speech Processing Toolkit
ukrainian-onnx-model - An ONNX model for speech recognition of the Ukrainian language
SincNet - SincNet is a neural architecture for efficiently processing raw audio samples.
NeMo - A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)
speech-to-text-benchmark - speech to text benchmark framework
Kaldi Speech Recognition Toolkit - kaldi-asr/kaldi is the official location of the Kaldi project.
imgaug - Image augmentation for machine learning experiments.
denoiser - Real Time Speech Enhancement in the Waveform Domain (Interspeech 2020)We provide a PyTorch implementation of the paper Real Time Speech Enhancement in the Waveform Domain. In which, we present a causal speech enhancement model working on the raw waveform that runs in real-time on a laptop CPU. The proposed model is based on an encoder-decoder architecture with skip-connections. It is optimized on both time and frequency domains, using multiple loss functions. Empirical evidence shows that it is capable of removing various kinds of background noise including stationary and non-stationary noises, as well as room reverb. Additionally, we suggest a set of data augmentation techniques applied directly on the raw waveform which further improve model performance and its generalization abilities.
best-of-ml-python - 🏆 A ranked list of awesome machine learning Python libraries. Updated weekly.