AI/ML●●Solid
Speechos – Benchmark 25 speech AI models locally, no cloud needed
Side-by-side model comparison eliminates guessing which speech engine fits your hardware.
Dark HorseSolve My Problem
hamuf
113mo ago
Back to browse
GitHub Repository
A lightweight library for normalizing speech transcripts before computing WER
26 starsPython
Reproducible open-source STT API benchmarks with full methodology
by jilijeanlouis·Mar 24, 2026·1 point·1 comment
AI Analysis
●●SolidSolve My ProblemCozy
Fixes WER scores by normalizing '$50' and 'fifty dollars' as equivalent.
Strengths
- •Three-stage pipeline (pre-process, word, post-process) is deterministic and YAML-configurable.
- •Expands contractions, converts symbols, removes fillers before WER computation.
- •Language-aware normalization prevents false penalties for formatting differences.
Weaknesses
- •From Gladia—an STT vendor—may bias toward their evaluation needs.
- •Only 2 GitHub stars suggests limited community adoption so far.
Category
Target Audience
STT engineers, ML researchers evaluating speech models
Similar To
jiwer · SpeechBrain · Kaldi scoring tools
Similar Projects
Security●●Solid
Teapot – A methodology for pen testing voice AI agents
Voice-specific prompt injection framework, but testing methodology alone isn't a shipping product.
Big BrainNiche Gem
xmhatx
7103mo ago
AI/ML●●Solid
Reproducible benchmark – OpenAI charges 1.5x-3.3x more for non-English
Exposes 230% Arabic token tax that nobody talks about in pricing.
Dark HorseBig Brain
vfalbor
101mo ago
Developer Tools●●Solid
Open-source benchmark for transcription APIs on meeting audio
Tests Deepgram and AssemblyAI on actual crosstalk instead of clean audiobook samples.
Solve My ProblemNiche Gem
eyepaqio
202mo ago
Productivity●●●Banger
Migas – Meeting copilot with live speaker labels (no bot, no cloud STT)
On-device speaker ID without training beats Granola and Otter's cloud-dependent approach.
WizardrySolve My ProblemZero to One
blakers95
122mo ago
