whisper-base-id

openai/whisper-base (74M params) fine-tuned for Indonesian: the accuracy/speed middle ground, realtime on Cortex-A53 (with ~1.9 s latency) and ~0.2 s on RK3588. Code: github.com/porcupine-md/whisper-id. Siblings: whisper-tiny-id (lighter), whisper-small-id (more accurate).

Training: 284.8 h (Common Voice 17 id, FLEURS id_id, YODAS2 id000 filtered with large-v3-turbo, YODAS labels = turbo transcripts), 6000 steps; telephone / Opus / AAC / noise augmentation; 80% of batches with short encoder context (mel cut to the clip, as whisper.cpp -ac).

Results

WER / CER (%), full test sets, lowercase + punctuation stripped:

Latency, whisper.cpp q8_0, 4 s utterance:

Output is lowercase without punctuation.

Files

• model.safetensors (fp16) + tokenizer/config: transformers
• ggml-base-id-q8_0.bin (82 MB): whisper.cpp. On ARM, q8_0 beats q5_0 on speed.

Usage

# -ac = encoder positions: 50 per second of audio, round up to a multiple of 64 (4 s -> 256, 8 s -> 448)
whisper-cli -m ggml-base-id-q8_0.bin -l id -t 4 -ac 256 -bs 1 -bo 1 -nt -f audio16k.wav

from transformers import pipeline
asr = pipeline("automatic-speech-recognition", model="maleo-ai/whisper-base-id")
print(asr("audio.wav", generate_kwargs={"language": "indonesian", "task": "transcribe"})["text"])

Limitations

• Test sets are read speech; conversational audio will be harder.
• Orange Pi Zero3 has no thermal throttling (only a 110 C shutdown); sustained 4-thread decoding runs above 80 C. Use a heatsink.

Citation

@misc{irawan2026whisperbaseid,
  author       = {Eka Tresna Irawan},
  title        = {whisper-base-id: Whisper base fine-tuned for Indonesian speech recognition},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/maleo-ai/whisper-base-id}}
}