Measured modelsVoxpopuli-AA-Cleaned WER
Clear benchmark ×Parakeet TDT 0.6B v3 is NVIDIA's 600M-parameter multilingual speech-to-text model built on the FastConformer-TDT architecture. Trained on the Granary dataset (670,000+ hours of audio), it supports automatic…
1B parameters · 25 languages · Encoder: Conformer-based · Decoder: Transformer
2.5B parameters · Encoder: Conformer-based · Decoder: Qwen2
0.64B parameters · 40 languages · Encoder: FastConformer (cache-aware) · Decoder: RNNT
0.62B parameters · Encoder: FastConformer (cache-aware) · Decoder: RNNT
1.1B parameters · Encoder: Conformer-based · Decoder: CTC
0.6B parameters · Encoder: Conformer-based · Decoder: TDT / RNN-T
0.6B parameters · Encoder: Conformer-based · Decoder: TDT
0.6B parameters · 26 languages · Encoder: Conformer-based · Decoder: TDT