NepTel 0.2

Real Nepali phone audio for evaluating speech recognition: 78 segments, 2,647 words, 31.5 minutes, 8 kHz. 56 segments are customer-support calls from the InfoBay public samples and 22 are new calls with our voice assistant. The audio is ungated: no login and no access request.

References. Every reference was listened to and corrected by a human. Corrections started from a draft by our LONG1C model, so LONG1C is unranked. The v0.1 references, drafted by Google Chirp 2 and then reviewed, are kept below for the record.

Evaluation only. Do not train on this data. Canary: NEPTEL-CANARY-2026-8f3a1c92

Files are in v0.2/: audio/ and references.json.

system	mistakes per 100 words	gap to Rani (95% CI)
Rani, no language model	33.5
IndicConformer Finetune + NepaliConformer Distilled (KritiTelephony)	37.9	+4.4 (+1.9, +6.9)
NepaliConformer offline	39.7	+6.2 (+3.2, +9.0)
AI4Bharat IndicConformer	47.6	+14.1 (+10.4, +18.0)

Score a system

git clone https://github.com/Ampixa/rani-asr
cd rani-asr/benchmark
python fetch_audio.py neptel_audio        # pulls v0.2/audio from this dataset, no login

python run_system.py --hf openai/whisper-large-v3 --name whisper --lang ne
python run_system.py --nemo your_model.nemo --name mine
python run_system.py --cmd "your-cli -f {audio}" --name mine
python run_system.py --py yourpkg.module:transcribe --name mine

Each writes outputs/<name>.json and prints the score. Add --compare outputs/rani.json for a paired bootstrap against Rani.

---

NepTel v0.1 (historical)

75 scored segments / 2,375 reference words of real Nepali call-center audio (genuine two-party customer-support calls), with human-reviewed reference transcripts. To our knowledge this is the first public Nepali ASR benchmark on real call audio rather than read-aloud speech.

This repository exists so anyone can benchmark a Nepali ASR system without any access request: the audio is cut and ready, no gate, no approval step.

Evaluation only. Do not train on this data. Canary: NEPTEL-CANARY-2026-8f3a1c92

Contents

• audio/ — 77 segment wavs (22.05 kHz mono, ~31 minutes total)
• references.json — reference transcripts and per-segment provenance

references.json lists all 77 segments; the 2 carrying an excluded field (one caught by the speaking-rate gate, one flagged unintelligible in native-speaker review) are skipped by the scorer, leaving the 75 scored segments / 2,375 words quoted in published results. They are kept rather than deleted so the exclusions stay auditable.

Standings at v0.1

See the v0.1 results.

Reference protocol

1. Audio cut into fixed 25-second windows from the vendor's dual-channel per-speaker files.
2. Reference drafts by Google Chirp 2 (ne-NP).
3. A native Nepali speaker reviewed every segment against the audio: accept / correct / flag. Flagged segments are excluded. 95.9% of usable segments in the most recent batch were accepted verbatim.
4. A speaking-rate gate (>6 words/sec) guards against transcription-engine hallucinations becoming ground truth.

Known limitations: 3 calls from one vendor; vendor-side PII muting leaves digital-zero gaps in some segments; references are reviewed drafts rather than from-scratch transcriptions; systems trained on Chirp 2 pseudo-labels (including ours) share label lineage with the reference drafts. Sized for ±2-point deltas between systems, not for absolute-truth WERs. Full provenance, including declared conflicts of interest, is in PROVENANCE.md.

Source audio and license

Segments are cut from InfoBayAI/Nepali_Call_Center_Audio_Dataset_Dual_Channel, published by InfoBayAI under CC-BY-4.0, which permits redistribution with attribution. The audio here is that dataset, cut into segments; the reference transcripts are ours, also CC-BY-4.0. Please credit both InfoBayAI (source audio) and Ampixa Labs (segmentation and references).