changed asr models outputs to be consistent

[Ssofja]

What does this PR do ?

This PR is making all ASR models outputs to be consistent and always user can expect List[Hypothesis] by default.

Collection: ASR

Changelog

• Changed ASR Models (RNNT, CTC,RNNT/CTC hybrid and AED) outputs.
  • List[Hypothesis]
  • If the type of hypotheses is NBestHypotheses, models return List[List[Hypothesis]]
• Renamed Timestep to Timestamp in the case of these models
• Added some changes in codes based on the first bullet point

Usage

From command-line

with transcribe_speech.py script

python transcribe_speech.py pretrained_name="model_name" \
dataset_manifest=<manifest_path> \
output_filename=<output_filename> 

From Python Env

The change is done for CTC, RNNT and AED models

Previous version

Where best_hypotheses is List[str], all_hypotheses is List[List[str]]
And in the case when return_hypotheses is True it returns a tuple of List[Hypothesis] and Optional(List[List[Hypothesis]])

from nemo.collections.asr.models import ASRModel
model = ASRModel.from_pretrained('<model_name>')
best_hypotheses, all_hypotheses = model.transcribe(['<file_path>'], return_hypotheses=False)

New version

where hypotheses is List[Hypothesis] or if hypothesis type is NBestHypotheses, the function will return List[List[Hypothesis]]. In hypotheses here only y_sequence, score, text variables are set, all other variables are default values
If return_hypotheses is True all variables of hypotheses are set

from nemo.collections.asr.models import ASRModel
model = ASRModel.from_pretrained('<model_name>')
hypotheses = model.transcribe(['<file_path>'], return_hypotheses=False) 

Each output contains hypothesis object and user can get text, timestamp and other arguments from Hypothesis data class:

It is important to remember, that in previous versions transcribe method had an argument timestep which was renamed to timestamp

output = model.transcribe(['<file_path>']) 
text = output[0].text
timestamp = output[0].timestamp #-> but this returns [] as timestamps were not requested

#Another case is 

output = model.transcribe(['<file_path>'], timestamps=True) 
text = output[0].text
timestamp = output[0].timestamp

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

PR Type:

• New Feature
• Bugfix
• Documentation
• Refactoring

[nithinraok]

Overall looks good! Thank you.
Changes required:

1. rename return_hypotheses -> return_all_hypotheses
2. move from draft to ready
3. Make sure all variables from Hypotheses return None if not set previously using return_all_hypotheses
4. update timestep to timestamp in Hypothesis object (used with timestamps=True feature)
5. Update docs with this feature update
6. Update PR changelog to explain changes and usage as similar to Timestamps to transcribe #10950

[nithinraok]

Few comments, otherwise LGTM

[titu1994]

I've only glanced at it but this PR is full of breaking changes. Upto @nithinraok if he's ok with it, I don't see the reason for most of these naming changes.

[github-actions]

[🤖]: Hi @Ssofja 👋,

We wanted to let you know that a CICD pipeline for this PR just finished successfully

So it might be time to merge this PR or get some approvals

I'm just a bot so I'll leave it you what to do next.

//cc @pablo-garay @ko3n1g

[nithinraok]

LGTM. Please remember to update:

1. huggingface model cards and spaces using main
2. Add to docs