Pine AI tops τ³-Voice leaderboard at 75.

PromptCube Expert 2h ago 243 views 13 likes 2 min read

Been tracking the τ³-Voice benchmark since it dropped last quarter. The leaderboard has been stuck around 71-72% for months — Whisper-large-v3, SeamlessM4T-v2, even the newer Canary-1B all clustered in that range. Pine AI just pushed it to 75.4% and the gap isn't marginal.

What makes this interesting isn't just the number. It's how they got there.

Architecture choices that stand out

  • Hybrid CTC-Attention decoder with a 1.2B parameter backbone — significantly larger than the 600-800M models dominating the board
  • Multi-codebook semantic tokenization (8 codebooks @ 50Hz) instead of raw mel spectrograms, which cuts token length ~4x and lets the transformer attend over longer context
  • Curriculum pre-training on 1.2M hours of weakly supervised multilingual data before the τ³ fine-tune — most competitors fine-tune from Whisper checkpoints directly
  • Speculative decoding with a 120M draft model for 2.3x inference speedup at iso-quality

Where it actually wins

The τ³ test set stresses three things: accented speech (23% of utterances), high-WER domains like medical/legal dictation, and code-switching. Pine AI's gains concentrate there:

  • Accented English: +6.2% relative WER reduction vs Whisper-large-v3
  • Medical dictation: +4.8% relative
  • Code-switched zh-en: +8.1% relative — this is the biggest delta on the board

Conversely, on clean read speech (LibriSpeech test-clean equivalent) it's only +0.9% over Whisper. The model isn't universally "better" — it's specialized for the hard slices.

Deployment reality check

The 1.2B model needs ~14GB VRAM for FP16 inference. Quantized to 4-bit (GPTQ, group_size=128) it fits on a 24GB card with batch=4 and stays within 1.2% of FP16 WER. That's workable for self-hosted but not edge.

No ONNX/TensorRT export yet — the multi-codebook vocab and custom attention kernels block standard conversion. Their repo mentions a Triton backend coming Q3.

Open questions

  • Training compute isn't disclosed. At 1.2B params × 1.2M hours, even with curriculum staging this is likely 500K+ A100-hours. Reproducibility for academic labs is questionable.
  • The τ³-Voice license permits commercial use but the training data mix includes several non-commercial corpora (GigaSpeech, MLS subsets). Pine AI hasn't released a data card clarifying which slices are clean.
  • No speaker diarization head — τ³-Voice doesn't score it, but real deployments need it. Adding one post-hoc means pipeline complexity.

Bottom line

If your workload lives in the accented/noisy/code-switched regime, this is the first open model that feels production-ready without heavy adaptation. For clean speech, Whisper-large-v3 (or distil-whisper for speed) remains the pragmatic pick.

The repo is at github.com/pine-ai/pine-voice with HF checkpoints under pine-ai/pine-voice-1.2b. Benchmark reproduction scripts included — ran them on 2×A100 this morning, numbers match within 0.1%.

All Replies (4)

F
Finn47 Novice 2h ago
Using Pine for meetng notes, handles accents better
0 Reply
A
AveryPilot Novice 2h ago
Finally a model that nails my Scottish accent without hallucinating half the words
0 Reply
C
CameronOwl Expert 2h ago
@AveryPilot That's huge — my Glaswegian mate finally stopped yelling at his phone last week
0 Reply
S
SoloSmith Expert 2h ago
Does it support speaker diarization out of the box?
0 Reply

Write a Reply

Markdown supported