Downloads · 30 days
15
47% of all-time downloads
hirotakahiraki/parakeet-turntaking-stage2-synth-3tag-ttconf90-down
parakeet-turntaking-stage2-synth-3tag-ttconf90-down is a machine learning model from hirotakahiraki. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Speech-LLM checkpoint for turn-taking prediction, trained on synthesized audio with cross-annotation labels. Step 3000.
Downloads · 30 days
15
47% of all-time downloads
All-time downloads
32
Public
Repo size
6 GB
Likes
0
Public
Click a slice to open those files.
.pt6 GB · 100%
From the Hugging Face model README
Speech-LLM checkpoint for turn-taking prediction, trained on synthesized audio with cross-annotation labels. Step 3000.
| Tag | Description |
|---|---|
<COMPLETE> | Utterance is semantically complete |
<INCOMPLETE> | Utterance is semantically incomplete |
<BACKCHANNEL> | Brief acknowledgment |
| Component | Model | Parameters |
|---|---|---|
| Encoder | nvidia/parakeet-ctc-0.6b (FastConformer CTC) | 600M |
| Projector | Conv1d + Transformer (1024 → 896, stride=2) | trainable |
| LLM | Qwen/Qwen2.5-0.5B | 500M |
llm_and_projector — LLM and projector frozen, encoder frozeninteract_canary_wordlevel (step 70,000)This Stage2 model is initialized from a Stage1 ASR-alignment checkpoint.
stage1-interact-canary-wordlevel in project parakeet-turntakingall_trainablemean_prob >= 0.9stage1_librispeech_spk step 90,000 (LibriSpeech 960h with speaker kernel)| Split | Samples |
|---|---|
| Train | 847,835 |
| Dev | 10,156 |
| Test | 12,261 |
Source: CANDOR corpus — naturalistic English dyadic conversations.
Audio is voice-cloned synthesized speech from CANDOR transcripts using Canary TTS, verified to have WER=0 against the original transcript. Turn-taking labels are derived from a cross-annotation pipeline (LLM + TEN confidence scoring).
Data filters applied:
tt_confidence >= 0.9 for COMPLETE and INCOMPLETE labels (high TEN agreement)Label distribution:
| Split | COMPLETE | INCOMPLETE | BACKCHANNEL | Total |
|---|---|---|---|---|
| Train | 14,146 | 14,146 | 2,717 | 31,009 |
| Dev | 6,850 | 1,975 | 376 | 9,201 |
| Test | 8,323 | 2,232 | 439 | 10,994 |
Note: Downsampling (COMPLETE→INCOMPLETE ratio=1.0) is applied only to the train split.
import torch
checkpoint = torch.load("step_3000.pt", map_location="cpu")
# Load into SpeechLLM model — see the parakeet-turntaking repository for details
step_3000.pt — Full model checkpoint (encoder + projector + LLM state dicts)config.yaml — Training configuration