Build a Custom ASR Model in TensorFlow: A Step-by-Step Tutorial
Vložit
- čas přidán 26. 02. 2023
- Learn the basics of speech recognition with TensorFlow and build practical applications with this tutorial. Discover the history of speech recognition and the challenges that come with dealing with human speech variability, similar-sounding words, and low-quality audio. Explore the various techniques used in speech recognition, such as machine learning algorithms like deep learning, Hidden Markov Models (HMM), Dynamic Time Warping (DTW), and phonetic-based approaches. Discover how transformers have transformed the field of speech recognition and how they can be used to recognize different languages, understand natural language, and distinguish between similar words. Follow along with the tutorial to build a basic speech recognition model using TensorFlow, combining a 2D convolutional neural network (CNN), recurrent neural network (RNN), and Connectionist Temporal Classification (CTC), and apply this knowledge to develop practical applications.
Text Version Tutorial: pylessons.com/speech-recognition
GitHub: github.com/pythonlessons/mltu...
pypi: pypi.org/project/mltu/
#machinelearning #python #tensorflow #opencv #ASR
A good presentation. Thank you for providing this information.
Glad it was helpful!
This is so nice. Thank you very much for sharing your knowledge.
You are welcome!
thank you for efforts, after train and save model how i use to transcript other audio not the one i trained and exist on csv file ? please tell me ? another thing how i know train is good with curves.
I am looking for some resources to learn ASR but I couldnot find good resources so could you please share me some ASR resources. Thank You!
Will there be a PyTorch version of this tutorial??? It would be great. Thanks for such helpful video.
I am not sure if its necessary, I already made a pytorch tutorial for handwritten words recognition, its pretty easy to combine both of them to get same results
That's great, thanks for your sharing.
After creating the model, can we use this model with openai whisper ?
Hey, I don't know, never used openai whisper
@@PyLessons thanks a lot, have a nice day 😁
thank you for the nice tutorial I think you did it with CTC mode which is sequence to sequence. I want to do the same project by using my dataset by using Listen attend and spell model and there is no any tutorial done on that area can you help me on how to implement it??
There is plenty tutorial online, I don't have time to try this. Hope you understand me
Thanks.. Fantastic work.. Please can I run it in my own CPU computer??
I think you can't train it on cpu, but if only using it, then yes, you can
when i try your code , on the output folder model I did not get model.onnx file
and when i test .h model i get error message said "model,onnx not found"
can you help me ?
This means there is something wrong with onnx package on your side. Check in terminal, there should be an exception what is wrong
why you select 1000 as epochs number ?
I select 1000, not to make any limitations. I am using callbacks that will break the training process when model stops improving
nice explaination but please can you add a method in which user can recognize his own voice by repeating dataset sentences
You will need to do this by your self, I am giving a principle how it works
Could you please make video on project converting text to speech ?
Maybe in the future, its way more difficult to convert text to speech, you should use transformers for this task
@@PyLessons What kind of transformers?
can i use this for making a model for arabic language ?
You can try, I am not sure how it will result
why dont you put microphone on your model? i just wonder
can you provide your pretrained model for use as we cannot train on cpu
Why you need it then if you can't train it, because I demonstrated it simply as example, not production or something ready model
can you please answer my questions ?
Answered it for you