AppTek Call-Center Dialogues (Manual Segmentation): leaderboard

AppTek Call-Center Dialogues (2026) is a long-form English speech recognition test set of spontaneous role-played call-center conversations in 14 accents, recorded for evaluation only. This board is the word error rate in percent, averaged over the accents, with manual segment boundaries; lower is better.

Metric: Word error rate (%, averaged over the 14 accents) on AppTek Call-Center Dialogues (128.6 hours of spontaneous role-played English call-center conversations, 14 accents), default inference settings, scored per session after Open ASR Leaderboard text normalization, audio segmented with manual (human) segment boundaries; lower is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen3-ASR-1.7B7.9
2canary-qwen-2.5B8.6
3parakeet-tdt-0.6B-v38.8
4Qwen3-ASR-0.6B8.9
5Phi-4 Multimodal Instruct9.2#4154
6parakeet-tdt-0.6B-v29.2
7granite-speech-3.3-8B10.5
8canary-1B-v210.6
9Whisper Large-v310.7
10granite-speech-3.3-2B10.8
11kyutai-stt-pytorch-stt-2.6B-en11.1
12whisper-large-v218.5

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=apptek-call-center-dialogues-manual-segmentation · How It Works · Data refreshed daily, snapshot 2026-11-21.