We scope custom speech collection and annotation for AI training. Each agreed project defines the target languages, speaker profiles, dialects, recording format, metadata, background conditions, QA, rights, and acceptance criteria before production.
Spirelight is a speech data company. A project can cover speaker recruitment, custom recording, transcription, labeling, QA, rights, and delivery under one written scope. Buyers use custom work when an existing dataset does not fit a required language, accent, recording condition, label schema, or acceptance test.
Our work splits into two core services. Data collection captures new audio to your spec, from scripted prompts to spontaneous, far-field, and in-car speech in the languages and regional varieties required by the project. Annotation turns raw recordings into structured, labeled training data with transcripts, timestamps, speaker tags, and events. Recruitment, consent, recording, annotation, QA, rights, and delivery artifacts are defined for each project. Use the catalog to review language-specific collection configurations and planning inputs.
New to sourcing speech data? Start with our guides on what speech data is, how much you need, what audio annotation involves, speaker diarization and speaker labeling, data license agreements, and how to buy AI training data. Comparing vendors or scoping a project? See how to choose an audio annotation company and how a custom speech data collection project is scoped and run.
Commission a custom collection, have your audio annotated, or start from a language-specific collection configuration.
Custom scripted, spontaneous, wake-word, far-field, and in-car recording briefs, on-site in controlled environments or remote, with target speakers, demographic quotas, consent requirements, output structure, and acceptance criteria defined per project.
Transcription, speaker and event labeling, word-level timestamps, and intent and emotion tags on your audio, with the QA workflow and acceptance criteria agreed for each project.
Browse language-specific custom collection configurations with planned formats and project-specific terms. A rate, capacity, or speaker target appears only when that field has a dated evidence record; sample status, feasibility, rights, schedule, and final price are confirmed against your brief.
These custom-scoped offers combine collection, consent, annotation, and evaluation requirements for specific speech AI applications.
Private held-out test sets scoped to your callers, channels, languages, accents, noise conditions, and failure modes, with human-reviewed references and documented holdout rules.
Purpose-recorded, consented customer-service simulations with telephony delivery, separate-channel options, human-reviewed transcripts, and custom domain scenarios.
Separate-channel conversations that preserve natural overlap, interruptions, backchannels, repairs, and turn timing for speech-to-speech and low-latency voice systems.
Synchronized audio and video for visual speech, lip alignment, audiovisual agents, and digital-human research, with the intended model use stated in participant releases.
Short commands, numbers, item codes, confirmations, and corrections recorded for the headsets, languages, and operating noise used on the floor.
Consented simulations—not real victim calls—covering multilingual scenarios, urgency, interruptions, degraded phone audio, and human-reviewed reference transcripts.
Per-turn emotion labels on speech: discrete classes, optional valence and arousal ratings, intent kept separate, and per-class annotator agreement reported.
Turn-level emotion and sentiment labels on call audio for quality monitoring, escalation prediction, and agent assist, with a documented consent chain.
Turn-level affect labels on spontaneous conversational speech, so a voice agent can detect frustration and be evaluated on how it responds.
Recruited listener panels scoring mean opinion score, naturalness, and emotional appropriateness for expressive speech synthesis, with per-rater agreement.
Custom video datasets recorded to spec: talking-head and dialogue footage, expression and gesture sets, liveness sequences, and in-cabin captures, with per-participant consent.
On-camera monologue and dialogue for avatars, lip sync, and digital humans: fixed framing, frame-accurate sync, and likeness releases that name synthetic-media use.
Genuine and spoofed speech pairs for deepfake detector training: TTS, voice conversion, and replay attacks generated only from speakers who consented to the spoof.
Native contributor teams matched to the language and regional variant in the recording, with the written convention agreed before production starts. Start from the full transcription service overview, or see how the global contributor workforce behind every project is recruited and managed.
Annotators staffed by the recorded variant, from Castilian to Rioplatense, with the output convention fixed before the first file.
Coverage from Bavarian to Low German, with an orthographic convention agreed before production starts.
Neapolitan, Sicilian, and Venetian material handled as separate regional languages, not as accented Italian.
Seoul, Gyeongsang, and Jeolla audio, with the speaker's actual speech level preserved throughout.
Mandarin transcription with the character set, segmentation, and number conventions fixed upfront.
Hong Kong audio delivered as written Cantonese or Standard Written Chinese, in traditional characters, with code-mixed English kept.
A written script policy across kanji, hiragana, and katakana, agreed before the first file.
Dialect coverage across Egypt, the Gulf, the Levant, and Morocco, staffed by market.
France and francophone Africa staffed separately, with code-switching handled by convention.
Brazil and Portugal staffed separately, because the two markets do not share one annotator pool.
Sourced across Russian-speaking markets, with a stated convention for mixed contact speech.
Mixed English and Hindi speech transcribed to an agreed script policy, not treated as noise.
Loanword conventions and dialect assignment settled before the first file is processed.
Send over your speaker profiles, language needs, and background noise conditions. Our team will assess feasibility and identify what is needed to scope a written project proposal.
