Machine Learning / AI

How will regular semantic search behave in a support knowledge base if all documents contain positive recommendations, but the user asks what should definitely not be done?

How specifically to apply VAD to obtain quality samples?

How is structured output implemented as constrained decoding?

In TTS pipeline, training has become three times slower, GPU load 20%, GPU memory 90%. How to find the problem?

Alignment score и WER — это одно и то же?

Can changes in inference cause missing words and phrase repeats, and in which part?

MOS increased from 4.1 to 4.3, but the model sometimes misses words and repeats phrases. What to do?

Why might someone choose model A with WER 3% over model B with WER 2%?

What does WER = 0.05 mean?

# Complete this method def __init__(self, nums: List[float]): pass # Complete this method def dot_product(self, vec: 'SparseVector') -> float: pass

How to evaluate the quality of a TTS model?

Is it necessary to perform text normalization during preprocessing or inference?

Is it always necessary to convert text to phonemes?

If PCM is recorded at 8 kHz, but the WAV header indicates 48 kHz, what will happen and how will it sound?

Tell us about your experience working with TTS.

How to prepare 5000 hours of speech data of unknown quality for training TTS?

Did you conduct A/B tests independently or together with analytics?

Middle

ВкусВилл

What are context managers in Python, what are they used for, and where are they used?

Чем DDP отличается от DataParallel?

What is gradient boosting and how does it work?