MediaSpeech: Multilanguage ASR Benchmark and Dataset
The performance of automated speech recognition (ASR) systems is well known to differ for varied application domains. At the same time, vendors and research groups typically report ASR quality results either for limited use simplistic domains (audiobooks, TED talks), or proprietary datasets. To fill this gap, we provide an open-source 10-hour ASR system evaluation dataset NTR MediaSpeech for 4 languages: Spanish, French, Turkish and Arabic. The dataset was collected from the official youtube channels of media in the respective languages, and manually transcribed. We estimate that the WER of the dataset is under 5%. We have benchmarked many ASR systems available both commercially and freely, and provide the benchmark results. We also open-source baseline QuartzNet models for each language.
PDF AbstractCode
Datasets
Common Voice
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Speech Recognition | MediaSpeech | Quartznet | WER for Arabic | 0.1300 | # 1 | |
| WER for French | 0.1915 | # 3 | ||||
| WER for Turkish | 0.1422 | # 2 | ||||
| WER for Spanish | 0.1826 | # 3 | ||||
| Speech Recognition | MediaSpeech | wav2vec | WER for Arabic | 0.9596 | # 6 | |
| WER for French | 0.3113 | # 6 | ||||
| WER for Turkish | 0.5812 | # 6 | ||||
| WER for Spanish | 0.2469 | # 6 | ||||
| Speech Recognition | MediaSpeech | Deepspeech | WER for French | 0.4741 | # 7 | |
| WER for Spanish | 0.4236 | # 8 | ||||
| Speech Recognition | MediaSpeech | Wit | WER for Arabic | 0.2333 | # 2 | |
| WER for French | 0.1759 | # 2 | ||||
| WER for Turkish | 0.0768 | # 1 | ||||
| WER for Spanish | 0.0879 | # 1 | ||||
| Speech Recognition | MediaSpeech | Silero | WER for Spanish | 0.3070 | # 7 | |
| Speech Recognition | MediaSpeech | VOSK | WER for Arabic | 0.3085 | # 4 | |
| WER for French | 0.2111 | # 4 | ||||
| WER for Turkish | 0.3050 | # 5 | ||||
| WER for Spanish | 0.1970 | # 4 | ||||
| Speech Recognition | MediaSpeech | WER for Arabic | 0.4464 | # 5 | ||
| WER for French | 0.2385 | # 5 | ||||
| WER for Turkish | 0.2707 | # 4 | ||||
| WER for Spanish | 0.2176 | # 5 | ||||
| Speech Recognition | MediaSpeech | Azure | WER for Arabic | 0.3016 | # 3 | |
| WER for French | 0.1683 | # 1 | ||||
| WER for Turkish | 0.2296 | # 3 | ||||
| WER for Spanish | 0.1296 | # 2 |
