Paprika 🌶️ — Lithuanian Whisper (large-v3-turbo)

The most accurate freely available Lithuanian ASR we measured as of August 2026: 16.98% WER / 10.01% CER on a held-out VoxPopuli LT benchmark (11 plenary speeches, normalized text).

model	WER %
paprika (this model)	16.98
svogunas/whisper-large-v3-turbo-lt (base)	18.13
sam8000/whisper-large-v3-turbo-lithuanian…	26.75
openai/whisper-large-v3-turbo (stock)	30.03

Continuing the fine Lithuanian tradition of vegetable-named speech models: paprika is warm-started from svogunas 🧅 and fine-tuned for 8,000 steps (effective batch 32) on ~320 h of Lithuanian speech: LIEPA-3 spontaneous (245 h) + LIEPA-3 read (76 h) + VoxPopuli LT. Benchmark speakers were excluded from training. RTF ≈ 0.10 with MLX on Apple Silicon, ≈ 0.27 with HF transformers fp16 on MPS.

Public benchmark (FLEURS-lt, same-harness comparison)

Measured 2026-08-07 on identical hardware/protocol (greedy, language=lt, whisper-style normalization, FLEURS-lt test n=986):

model	WER
paprika (this model)	12.32 %
svogunas/whisper-large-v3-turbo-lt (base)	12.83 %

(The svogunas model card reports 13.18 % on its own harness — protocol differences account for the offset; the same-harness pair above is the fair comparison.) Our fine-tune improves on its base model out-of-domain as well as on the VoxPopuli bench.

Usage

from transformers import pipeline
asr = pipeline("automatic-speech-recognition",
               model="kristijonas/paprika-whisper-lt",
               chunk_length_s=30)
print(asr("audio.wav",
          generate_kwargs={"language": "lithuanian", "task": "transcribe"})["text"])

Known limitations (v1 — honest by design)

No punctuation or casing: output is lowercase and unpunctuated — the training labels lacked both. Pair with a punctuation/truecase post-processor for readable text.
Chunked inference only: native long-form/timestamp decoding degrades badly (the fine-tune lost timestamp competence). Use chunk_length_s=30 pipelines as in the example above.
16 kHz audio; parliamentary and spontaneous speech dominate the training mix.

A v2 addressing all three — timestamp-preserving recipe, punctuated labels, the full 10,000 h LIEPA-3, and a full large-v3 candidate — is in training. Issues, feedback, and hard Lithuanian test cases are very welcome.

Attribution

Base model: svogunas/whisper-large-v3-turbo-lt (CC BY 4.0)
Data: LIEPA-3 garsynas (CC BY 4.0, via raštija.lt / Vilnius University) — ačiū už atvirus duomenis; VoxPopuli (Meta AI)
Built by the Kalamo team — real-time translation and speech tools for smaller languages.