wav2vec2-xls-r-1b-italian-robust Open-source Speech Model - Free Deployment for Automatic Italian Speech Recognition

Wav2vec2 Xls R 1b Italian Robust

Developed by dbdmg

An Italian automatic speech recognition model fine-tuned on Common Voice 7 and Libri Speech datasets based on facebook/wav2vec2-xls-r-1b

Speech Recognition

Transformers

OtherOpen Source License:Apache-2.0 #Italian speech recognition #Large-scale speech model #Multi-scenario robustness

Downloads 130

Release Time : 3/2/2022

Model Overview

This is an automatic speech recognition (ASR) model optimized for Italian, based on the XLS-R architecture and fine-tuned on the Common Voice 7 and Libri Speech datasets.

Model Features

High-performance Italian recognition

Achieves 19.55% WER on the Common Voice 7 Italian test set (with language model)

Large-scale pre-training

Based on the 1-billion-parameter XLS-R architecture with powerful speech feature extraction capabilities

Language model enhancement

Supports language model integration, significantly improving recognition accuracy

Model Capabilities

Italian speech recognition

Speech-to-text

Long audio processing

Use Cases

Speech transcription

Italian speech-to-text

Convert Italian speech content into text format

Achieves 19.55% word error rate on Common Voice 7 test set

Voice assistants

Italian voice command recognition

Used for voice command recognition in Italian voice assistants or smart home systems

🚀 wav2vec2-xls-r-1b-italian-robust

This model is a fine - tuned version of facebook/wav2vec2-xls-r-1b on the Common Voice 7 & Libri Speech datasets. It is designed for automatic speech recognition tasks, offering improved performance in Italian speech recognition.

📚 Documentation

Model Information

Property	Details
Model Type	Fine - tuned version of facebook/wav2vec2-xls-r-1b
Training Data	Common Voice 7 & Libri Speech datasets
Tags	automatic-speech-recognition, generated_from_trainer, hf-asr-leaderboard, robust-speech-event
Datasets	mozilla-foundation/common_voice_7_0

Model Index

Name: XLS - R - 1b - Italian
Results:
- Task 1: Automatic Speech Recognition on Common Voice 7
  - Dataset: Common Voice 7 (mozilla-foundation/common_voice_7_0 with args it)
  - Metrics:
    - Test WER: 32.74
    - Test CER: 7.83
    - Test WER (+LM): 19.55
    - Test CER (+LM): 5.59
- Task 2: Automatic Speech Recognition on Robust Speech Event - Dev Data
  - Dataset: Robust Speech Event - Dev Data (speech-recognition-community-v2/dev_data with args it)
  - Metrics:
    - Test WER: 43.23
    - Test CER: 13.37
    - Test WER (+LM): 27.51
    - Test CER (+LM): 10.69
- Task 3: Automatic Speech Recognition on Robust Speech Event - Test Data
  - Dataset: Robust Speech Event - Test Data (speech-recognition-community-v2/eval_data with args it)
  - Metrics:
    - Test WER: 51.12

Evaluation Results

This model achieves the following results on the evaluation set:

Loss: 0.2428
Wer: 0.2960

🔧 Technical Details

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e - 05
train_batch_size: 32
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 10.0
mixed_precision_training: Native AMP

Training Results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.07	400	1.0053	0.8058
1.5087	0.13	800	0.9127	0.8104
0.9552	0.2	1200	1.0360	0.8836
0.9555	0.27	1600	0.9980	0.8577
1.0259	0.34	2000	1.0103	0.8842
1.0259	0.4	2400	0.9119	0.8466
1.0365	0.47	2800	0.9000	0.8281
1.0069	0.54	3200	0.7976	0.7875
0.9688	0.61	3600	0.8126	0.8051
0.9638	0.67	4000	0.7921	0.7903
0.9638	0.74	4400	0.7703	0.7783
0.9327	0.81	4800	0.7253	0.7463
0.8992	0.88	5200	0.6841	0.7171
0.8693	0.94	5600	0.6867	0.7250
0.8433	1.01	6000	0.7077	0.7302
0.8433	1.08	6400	0.6685	0.7091
0.8499	1.14	6800	0.6355	0.6825
0.8159	1.21	7200	0.6283	0.6800
0.8001	1.28	7600	0.6288	0.6743
0.7883	1.35	8000	0.5995	0.6633
0.7883	1.41	8400	0.6195	0.6726
0.7863	1.48	8800	0.6039	0.6588
0.7713	1.55	9200	0.5842	0.6490
0.7572	1.62	9600	0.5975	0.6533
0.7442	1.68	10000	0.5508	0.6233
0.7442	1.75	10400	0.5521	0.6209
0.7296	1.82	10800	0.5760	0.6245
0.7205	1.89	11200	0.5593	0.6144
0.7106	1.95	11600	0.5672	0.6220
0.7146	2.02	12000	0.5134	0.5911
0.7146	2.09	12400	0.5069	0.5811
0.6944	2.15	12800	0.5022	0.5962
0.6817	2.22	13200	0.4989	0.5813
0.6721	2.29	13600	0.4941	0.5742
0.6774	2.36	14000	0.4775	0.5676
0.6774	2.42	14400	0.4694	0.5525
0.6621	2.49	14800	0.4720	0.5514
0.6599	2.56	15200	0.4714	0.5553
0.6591	2.63	15600	0.4578	0.5397
0.645	2.69	16000	0.4619	0.5452
0.645	2.76	16400	0.4578	0.5343
0.6431	2.83	16800	0.4514	0.5328
0.636	2.9	17200	0.4526	0.5325
0.6433	2.96	17600	0.4561	0.5325
0.6356	3.03	18000	0.4386	0.5191
0.6356	3.1	18400	0.4291	0.5065
0.6175	3.16	18800	0.4306	0.5170
0.6187	3.23	19200	0.4256	0.5036
0.607	3.3	19600	0.4198	0.5027
0.6004	3.37	20000	0.4149	0.4906
0.6004	3.43	20400	0.4114	0.4902
0.6002	3.5	20800	0.4116	0.4967
0.5926	3.57	21200	0.4066	0.4843
0.5836	3.64	21600	0.3956	0.4791
0.588	3.7	22000	0.3941	0.4729
0.588	3.77	22400	0.3972	0.4799
0.5739	3.84	22800	0.4018	0.4790
0.5778	3.91	23200	0.3936	0.4750
0.5768	3.97	23600	0.3936	0.4751
0.5651	4.04	24000	0.3953	0.4706
0.5651	4.11	24400	0.3906	0.4659
0.5704	4.17	24800	0.3807	0.4557
0.5594	4.24	25200	0.3817	0.4610
0.5509	4.31	25600	0.3755	0.4553
0.5439	4.38	26000	0.3705	0.4471
0.5439	4.44	26400	0.3744	0.4487
0.5426	4.51	26800	0.3716	0.4483
0.5393	4.58	27200	0.3600	0.4356
0.5408	4.65	27600	0.3573	0.4307
0.5327	4.71	28000	0.3638	0.4382
0.5327	4.78	28400	0.3587	0.4316
0.5324	4.85	28800	0.3598	0.4290
0.5378	4.91	29200	0.3508	0.4243
0.5246	4.98	29600	0.3522	0.4260
0.5284	5.05	30000	0.3520	0.4268
0.5284	5.12	30400	0.3506	0.4224
0.5154	5.18	30800	0.3556	0.4223
0.5138	5.25	31200	0.3526	0.4276
0.51	5.32	31600	0.3440	0.4220
0.5065	5.39	32000	0.3367	0.4120
0.5065	5.45	32400	0.3406	0.4136
0.5087	5.52	32800	0.3370	0.4125
0.503	5.59	33200	0.3387	0.4134
0.5085	5.66	33600	0.3346	0.4068
0.5044	5.72	34000	0.3325	0.4057
0.5044	5.79	34400	0.3304	0.4026
0.4879	5.86	34800	0.3274	0.4002
0.4924	5.92	35200	0.3286	0.3980
0.4991	5.99	35600	0.3231	0.3952
0.487	6.06	36000	0.3324	0.4005
0.487	6.13	36400	0.3264	0.3952
0.4754	6.19	36800	0.3234	0.3905
0.4683	6.26	37200	0.3149	0.3840
0.4653	6.33	37600	0.3122	0.3824
0.4667	6.4	38000	0.3151	0.3855
0.4667	6.46	38400	0.3217	0.3859
0.4628	6.53	38800	0.3085	0.3831
0.4644	6.6	39200	0.3121	0.3791
0.4612	6.67	39600	0.3093	0.3790
0.4552	6.73	40000	0.3087	0.3749
0.4552	6.8	40400	0.3027	0.3679
0.4544	6.87	40800	0.3048	0.3672
0.4507	6.93	41200	0.2963	0.3614
0.4489	7.0	41600	0.3086	0.3718
0.4367	7.07	42000	0.3100	0.3754
0.4367	7.14	42400	0.3057	0.3701
0.4376	7.2	42800	0.2930	0.3614
0.428	7.27	43200	0.2907	0.3516
0.4241	7.34	43600	0.2916	0.3590
0.4312	7.41	44000	0.2904	0.3523
0.4312	7.47	44400	0.2908	0.3476
0.4292	7.54	44800	0.2858	0.3467
0.426	7.61	45200	0.2864	0.3484
0.4225	7.68	45600	0.2820	0.3441
0.422	7.74	46000	0.2834	0.3441
0.422	7.81	46400	0.2784	0.3420
0.4158	7.88	46800	0.2814	0.3390
0.4139	7.94	47200	0.2777	0.3384
0.4076	8.01	47600	0.2741	0.3381
0.3997	8.08	48000	0.2738	0.3320
0.3997	8.15	48400	0.2720	0.3303
0.4009	8.21	48800	0.2705	0.3357
0.3928	8.28	49200	0.2708	0.3265
0.3923	8.35	49600	0.2678	0.3283
0.3897	8.42	50000	0.2649	0.3241
0.3897	8.48	50400	0.2640	0.3218
0.3879	8.55	50800	0.2616	0.3197
0.3805	8.62	51200	0.2599	0.3170
0.3874	8.69	51600	0.2592	0.3168
0.3799	8.75	52000	0.2589	0.3157
0.3799	8.82	52400	0.2566	0.3137
0.3834	8.89	52800	0.2552	0.3141
0.3811	8.95	53200	0.2523	0.3108
0.3821	9.02	53600	0.2539	0.3112
0.3636	9.09	54000	0.2529	0.3070
0.3636	9.16	54400	0.2500	0.3078
0.3706	9.22	54800	0.2510	0.3067
0.367	9.29	55200	0.2497	0.3069
0.3618	9.36	55600	0.2493	0.3043
0.3624	9.43	56000	0.2491	0.3040
0.3624	9.49	56400	0.2466	0.3016
0.3557	9.56	56800	0.2460	0.3014
0.3536	9.63	57200	0.2470	0.2997
0.3584	9.7	57600	0.2441	0.2989
0.3563	9.76	58000	0.2442	0.2970
0.3563	9.83	58400	0.2436	0.2966
0.3492	9.9	58800	0.2431	0.2967
0.3483	9.96	59200	0.2428	0.2960

Framework Versions

Transformers 4.17.0.dev0
Pytorch 1.10.2+cu102
Datasets 1.18.3
Tokenizers 0.11.0

📄 License

This model is licensed under the Apache - 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご