Tokenize training transcripts by grapheme (cluster) instead of codepoint

Repository metrics

Stars

26,755 stars (26,755 stars)

PR merge metrics

No merged PRs in 30d (No merged PRs in 30d)

Description

This issue does not include a description.

Contributor guide

Research direction

Familiarize yourself with the current tokenization code in the DeepSpeech training pipeline (likely in Python). Study how transcripts are tokenized by codepoint. Then research how to tokenize by grapheme clusters using Unicode segmentation (e.g., the grapheme library in Python or ICU in C++). Identify the exact location in the codebase where tokenization occurs and propose changes to split by grapheme clusters instead of codepoints. Consider edge cases like combining characters and emoji sequences.

Tech stack

python

Domain

machine learningaidata

Issue type

Feature

DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.

3

Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.

Half day

Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.

Active

ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.

Clear

Prerequisites

PythonUnicode text processingDeepSpeech training pipeline

Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.

55