Identifying Speakers in Dialogue Transcripts: A Text-based Approach Using Pretrained Language Models
Paper • 2407.12094 • Published • 1
Joint SpeakerID MLX INT8
MLX-scaled INT8 weight-only quantisation of the Joint Speaker Identifier from adobe-research/speaker-identification (Interspeech 2024).
This checkpoint is a quantised child of Adobe Research’s original Joint Speaker Identifier (FP32). The architecture and trained weights are Adobe’s; Linear scales were computed with MLX and packed to INT8. No additional training.
| Metric | Value |
|---|---|
| Precision | 78.87 |
| Δ vs FP32 parent | 0.00 |
| F1 | 63.28 |
| Accuracy | 67.60 |
| Throughput | 18.16 examples/s (Apple M3 Pro, MPS, batch 2) |
| In-memory size | 472 MB (FP32 parent: 1633 MB) |
Matches FP32 Joint parent quality. Scales for Linear layers whose last dim is a multiple of 64 were computed with MLX (mx.quantize / mx.dequantize) and then packed for PyTorch/MPS inference. Smaller heads fall back to absmax INT8.
Parent model
| Parent | Joint Speaker Identifier (FP32), Adobe Research |
| Original weights | logs/mediasum-joint/best-model.mdl in adobe-research/speaker-identification |
| Paper | Identifying Speakers in Dialogue Transcripts: A Text-based Approach Using Pretrained Language Models (Interspeech 2024) |
| Backbone | FacebookAI/roberta-large |
| Relation | weight-only INT8 quantisation of the Adobe Joint checkpoint (MLX scales) |
Files
model.pt— quantised bundle (scheme=mlx_int8, group size 64)config.json— metrics and load metadata
Load
from huggingface_hub import hf_hub_download
path = hf_hub_download("hmarchant/speaker-id-joint-mlx-int8", "model.pt")
License
Derived from Adobe Research Speaker Identification. The Adobe Research License allows non-commercial research use only.
- Downloads last month
- 21
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for hmarchant/speaker-id-joint-mlx-int8
Base model
FacebookAI/roberta-large