100 hours of Bengali Scripted Speech

Speakers read prompts from a phonetically balanced script shown one line at a time, with retakes on any flubbed or misread line.

A 45-minute booth session yields roughly 300–400 prompts. One file per utterance, cut at the prompt boundary, so alignment is exact before any forced-alignment pass runs.

Yield at this style: 85–90% of recorded time is accepted. Prompt-level retakes catch problems inside the session, so very little is discarded afterwards. This is the highest-yield style we run.

• Script built for triphone coverage rather than word coverage, so rare phone contexts appear often enough to train on
• Digit strings, dates, currency and person names blocked separately, because these are where deployed ASR actually fails
• Domain lexicon injected from your product vocabulary when you supply one
• Sentence-length distribution spread deliberately, since all-short prompts produce a model that cannot handle long utterances
• Built against Bengali specifically: Inherent vowel is realised as /ɔ/ or /o/, which breaks G2P rules copied from Devanagari-based systems
• Code-mixing handled explicitly rather than edited out — Kolkata professional speech mixes English heavily; rural West Bengal much less. A single 'Bengali' dataset without register tags conflates two very different acoustic and lexical distributions.

Four to six weeks. Two cities running in parallel means recruitment and recording overlap rather than queue behind each other.

Three batches at roughly two-week intervals, each one a self-contained, speaker-disjoint slice you can evaluate independently.

Tag every speaker as Indian Bengali and record district of origin; mixing in Bangladeshi speech without tags is a common and costly dataset defect.

At 100 hours the cohort is deliberately simplified: two or three dialect groups rather than the full spread, with quotas enforced in aggregate. A build this size cannot support per-cell targets and should not claim to.

Requires literate speakers comfortable reading aloud in the target script, which is the main constraint on cohort breadth and has to be actively counterweighted.

• Inherent vowel is realised as /ɔ/ or /o/, which breaks G2P rules copied from Devanagari-based systems
• Kolkata professional speech mixes English heavily; rural West Bengal much less. A single 'Bengali' dataset without register tags conflates two very different acoustic and lexical distributions.
• Indian Bengali is under-collected relative to Bangladeshi Bengali, and North Bengal and Tripura varieties are almost entirely missing.

• Prompt-to-audio match verified by a native reviewer; a misread line is a rejected file, not an edited transcript
• Hyperarticulation flagged — a speaker over-enunciating because they are reading produces audio that does not match deployment
• Clipping and truncation checked at both utterance boundaries
• Per-speaker prompt coverage confirmed, so no speaker silently skips a block
• Three sibilant characters chosen inconsistently for the same sound
• Bangladeshi vs Indian Bengali orthographic conventions mixed within one dataset

100% technical QA, 25% content QA, escalating to full review on any batch that fails the agreed error threshold.

• Reading voice: flat prosody and unnatural stress that trains a model on speech nobody actually produces
• Prompt fatigue in the back half of long sessions, where accuracy drops and pace flattens
• Speakers who are not fluent readers, which quietly biases the cohort toward higher education bands
• Script leakage across speakers, producing a corpus that memorises sentences instead of covering sounds

Cost at this band is driven by: Dialect spread is the main driver at this band — two contrasting cities cost more than two convenient ones; Turnaround compression, if you need it inside four weeks.

• Two-city cohorts can hide a dialect gap that only appears when the model meets a third region
• Transcription consistency between two city teams needs an explicit convention document or the batches will not match
• Bengali carries 5 recognised varieties across West Bengal, Tripura, Assam (Barak Valley), so the quota matrix is wider than the headline volume suggests
• Indian Bengali is under-collected relative to Bangladeshi Bengali, and North Bengal and Tripura varieties are almost entirely missing.

• WAV audio to your naming convention, with the per-file manifest
• Verbatim Bengali transcripts with utterance-level timestamps
• Prompt ID mapped to every utterance
• Verbatim deviation flags where the speaker departed from the script
• Per-utterance SNR and duration in the manifest
• Per-speaker metadata: age band, gender, region, dialect, education band
• Consent records mapped to speaker IDs
• QA report with pass rates, rejection reasons and agreement statistics
• Speaker-disjoint train / dev / test splits on request

• ASR acoustic model baselines
• TTS voice building where a single speaker is recorded at depth
• Pronunciation lexicon and G2P validation
• Forced-alignment and phone-boundary work

Scripted audio contains no disfluencies, no false starts and no natural turn-taking. A model trained on it alone degrades sharply on real spontaneous input, so it is a baseline layer rather than a complete training set.