Learn Before
Transcribing Spoken Notes and the Data Needed for End-to-End Training
Case context: A company is building a tablet feature that turns recorded voice memos into written text. The team wants a single end-to-end model that takes an audio clip and outputs the full transcript.
Question: How should this output be classified, and what training data does the team need in order to fit the end-to-end model?
Sample answer: The output should be treated as a rich output because it is a full sentence or multiword transcript rather than a single scalar value. To train the model, the team needs the correct labeled input-output pairs, specifically audio recordings matched with their corresponding text transcripts.
Key points:
- The transcript is a rich output.
- Training requires labeled input-output pairs.
- The relevant pairs are audio clips and their transcripts.
Rubric: The response must classify the transcript as a rich output and state that training requires matched audio and transcript pairs.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
In an end-to-end speech recognition system, what goes in and what comes out?
A speech-to-text transcript is a rich output because it can contain many words rather than a single number.
Speech Recognition Inputs
Match each item in an end-to-end speech system to what it represents.
Order the checks for deciding whether a problem fits direct end-to-end learning with rich outputs.
What trend does modern deep learning show for end-to-end systems with complex outputs?
End-to-end learning can sometimes map inputs directly to complex outputs such as transcripts, masks, or other structured labels when suitable labeled pairs are available.
End-to-end systems that produce rich outputs are an _____ direction in deep learning.
Match each output example to whether it is a rich output or a single-number output.
Put the stages of an end-to-end speech recognizer into the correct order.
Why Speech-to-Text Is a Rich-Output Task
Transcribing Spoken Notes and the Data Needed for End-to-End Training
Input and Output in a Speech-to-Text System