Gipformer 1.5 - Efficient Vietnamese Speech Recognition

Source code: github.com/ggroup-ai-lab/gipformer

Highlights

State-of-the-art accuracy — Demonstrates top-tier performance across major Vietnamese ASR benchmarks, delivering highly precise and reliable transcription quality.
Robust handling of telephonic domains — Excels in processing challenging, noisy real-world call center recordings across all major Vietnamese regional accents.
Domain-leading accuracy — Best WER on the technology, finance, education, public administration and medical benchmarks.
Outstanding parameter efficiency — Ranks among the smallest ASR models currently available.
Seamless edge deployment — Its naturally low resource requirements enable ultra-fast inference on mobile and embedded systems, making it perfectly suited for offline, on-device applications.
Built-in data privacy — By supporting full local execution, the model ensures sensitive audio data is processed securely on-device, eliminating the need for third-party cloud services.
gipformer1.5-68M-rnnt is based on the Zipformer Transducer architecture.

Benchmark Results (WER%)

Lower is better; bold = best in each column.

Normalization: Both predictions and labels are normalized before computing WER — lowercased, punctuation removed, and numbers converted to spoken form.

Model	Params	tele-medium	tele-hard-north	tele-hard-middle	tele-hard-south	vi-asr-tech	vi-asr-edu	vi-asr-finance	vi-asr-pubadmin	vivos	Common-Voice	vlsp-t1	VietMed	MultiMED	LSVSC	Fleurs	ViMD
vinai/PhoWhisper-small	244M	33.96	55.88	65.41	62.35	56.81	51.87	47.36	41.46	6.23	11.55	15.99	25.50	26.02	11.23	16.11	14.09
vinai/PhoWhisper-medium	769M	26.46	51.20	59.04	55.39	47.93	43.00	41.60	34.81	4.93	8.37	14.06	24.90	24.76	10.25	14.44	11.34
vinai/PhoWhisper-large	1.5B	26.82	50.39	59.44	56.70	42.94	44.99	40.54	33.78	4.73	8.60	13.70	24.37	24.47	10.08	12.62	11.18
Qualcomm-AI-Research/PhoASR-whisper-small	244M	30.73	50.30	56.66	54.12	44.35	41.65	—	35.99	5.87	9.73	14.60	23.10	22.78	7.59	13.97	9.75
nvidia/parakeet-ctc-0.6b-Vietnamese	600M	31.82	55.33	61.65	56.70	48.69	45.86	43.16	38.84	7.76	11.40	17.00	23.53	23.79	10.46	16.11	12.95
khanhld/chunkformer-large-vie	110M	27.60	46.30	51.91	49.09	39.81	37.20	34.31	29.30	4.18	6.94	14.09	19.59	22.60	8.85	14.17	11.77
nguyenvulebinh/wav2vec2-base-vi	95M	23.71	40.49	48.90	46.33	59.86	53.06	49.66	39.82	6.60	12.61	13.14	22.96	23.03	9.89	20.09	11.42
zipformer-rnnt	68M	20.30	42.21	49.01	47.86	36.81	30.26	29.13	18.62	6.92	11.48	14.54	21.90	22.05	10.23	14.76	10.15
Qwen/Qwen3-ASR-1.7B	1.7B	26.34	46.80	59.85	51.84	27.95	29.93	34.09	31.83	7.17	10.76	16.29	20.21	20.11	9.64	10.13	11.16
Qwen/Qwen3-ASR-0.6B	600M	32.29	48.57	61.88	55.43	37.52	36.97	38.31	38.65	10.23	16.68	18.62	22.51	22.65	10.96	13.11	14.37
hynt/Zipformer-30M-RNNT-6000h	30M	19.53	38.13	44.73	41.58	29.77	25.91	25.86	17.92	4.55	4.16	11.78	19.91	19.88	9.04	13.03	7.18
g-group-ai-lab/gipformer-68M-rnnt	68M	15.53	25.10	32.27	32.62	36.59	29.81	29.67	20.09	4.12	6.63	13.39	19.41	19.35	8.96	12.92	7.17
g-group-ai-lab/gipformer1.5-68M-rnnt	68M	15.44	26.24	33.31	32.48	27.49	23.32	22.34	14.82	4.25	6.45	13.37	19.23	19.17	8.97	12.65	7.00

Dataset Descriptions

Private test sets (call center domain):

tele-medium — Call center recordings with medium difficulty
tele-hard-north — Low-quality call center audio, hard-to-hear speakers — Northern Vietnamese accent
tele-hard-middle — Low-quality call center audio, hard-to-hear speakers — Central Vietnamese accent
tele-hard-south — Low-quality call center audio, hard-to-hear speakers — Southern Vietnamese accent

Public test sets:

vi-asr-tech-test — Technology: consumer electronics reviews, software tutorials, programming and IT walkthroughs
vi-asr-edu-test — Education: study-abroad consulting, exam guidance, university and training-course introductions
vi-asr-finance-test — Finance: stock market commentary, trading platforms, banking and crypto
vi-asr-pubadmin-test — Public administration: administrative procedures, licensing guidance, civil records
vivos — Vietnamese read speech corpus
Common-Voice — Mozilla Common Voice, Vietnamese subset
vlsp-t1 — VLSP 2020 ASR Shared Task 1
VietMed — Vietnamese medical domain
MultiMED — Multi-domain medical conversations
LSVSC — Large-Scale Vietnamese Speech Corpus
Fleurs — Google's Few-shot Learning Evaluation of Universal Representations of Speech, Vietnamese subset
ViMD — Vietnamese Multi-Domain

Call Center Domain: Where It Matters Most

Call center ASR is one of the most challenging real-world domains — noisy phone lines, overlapping speech, diverse regional accents, and spontaneous conversation. The gipformer family holds the best result on all four call center test sets, and no other model comes within 9 WER points on any of the three hard subsets.

Usage

See the Quick Start guide for detailed usage instructions.

Citation

@misc{gipformer,
  title={Gipformer - Efficient Vietnamese Speech Recognition},
  author={G-Group AI Lab},
  year={2026},
  url={https://huggingface.co/g-group-ai-lab/gipformer1.5-68M-rnnt}
}

License

This model is released under the MIT License.

Acknowledgments

Developed by G-Group AI Lab. For questions, issues, or collaboration inquiries, please visit our HuggingFace organization page.

g-group-ai-lab/gipformer1.5-68M-rnnt