--- title: "Speech Data for ASR, TTS, Voice AI | 150+ Languages | YPAI" url: https://ypai.ai/speech-data/ description: "Speech collected, labelled and evaluated where it happens: 150+ languages, Nordic depth, speakers by dialect, setting and device, consent and QA per recording." source: "src/copy/routes (route /speech-data/)" --- # Same test, Norwegian training. 36% lower WER out of domain. > Speech collected, labelled and evaluated where it happens: 150+ languages, Nordic depth, speakers by dialect, setting and device, consent and QA per recording. Speech models fail where their training data never went. Dialect, spontaneous speech, code-switching. YPAI collects, labels and evaluates speech in 150+ languages, with particular depth in Nordic and low-resource languages. Speakers are recruited by dialect region, setting and device, and every recording ships with its consent record, metadata and QA trail. Built in Norway, on EEA infrastructure, under Norwegian jurisdiction. NB-Whisper large 6.6% vs OpenAI Whisper large-v3 10.4% · FLEURS Bokmål · Kummervold et al., Interspeech 2024 ## Four axes decide whether your speech data survives deployment You name the range your system has to hold. The corpus is built to span it, and every recording lands with its coverage recorded. - The microphone in the user's hand, not the one in the studio. A model trained on one tier degrades on the next. - The room, the vehicle, the street. Noise, reverberation and competing speech are specified, not hoped for. - Recruited by dialect region, age and background, so the long tail is represented rather than averaged away. - 150+ languages, Nordic and low-resource depth - Targeted capture of the speech a benchmark removes and production does not. Coverage is agreed before recording begins and delivered with the dataset. Where a required range cannot be covered, YPAI says so during scoping. ## A benchmark covers one corner. A corpus covers the grid Every cell is a named condition, device by setting. Public benchmarks cluster in the quiet corner: read speech, one microphone, one dialect. Your corpus is specified against the cells your users speak in, and the coverage ships with the data. Coverage is recorded per recording and delivered with the dataset. ## Hear it on your own voice Record three seconds somewhere quiet. The same recording is transcribed clean, then again with contact-centre, cabin or street noise mixed in at a level you choose. The words that change are what training data has to cover. Word error rate against the quiet recording Runs in this tab, on your device Example shown until you run it. Illustrative, no customer data ## Choose the work that matches the gap Six routes into the same work. Each names what it produces and where its depth lives, and every route delivers consent, metadata and a QA trail per recording. ### [The speech you need does not exist yet](https://ypai.ai/speech-data/language-coverage/) - YPAI designs the collection, recruits and qualifies the speakers, and runs the recording operation. Read, prompted, spontaneous, conversational, command or wake-word speech, remote or in studio, delivered against agreed speakers, conditions and checks. ### [One voice matters more than population coverage](https://ypai.ai/audio/voice-recording/) - Specialist voice recording defines casting, pronunciation, performance and recording conditions. Recorded voice files with model-use rights stated in writing; voice-cloning, buyout, exclusivity and resale rights need an explicit written grant. ### [You have audio, but the model cannot use it](https://ypai.ai/annotation/audio-speech-annotation-services/) - Labelled corpora pair source audio with an agreed schema and review standard: transcription, segmentation, diarisation, alignment and linguistic annotation, checked by native reviewers with the QA trail kept. ### [Locate the failure before adding data](https://ypai.ai/speech-data/evaluation-program/) - Evaluation packages pair a set, rubric and reviewer process, then separate results by language, dialect, device, setting or speaker group. The output can include an error taxonomy and a regression set your team can run again. ### [Could an existing dataset fit?](https://ypai.ai/audio/datasets/) - YPAI reviews an existing or partner-sourced dataset against intended use, speech type, acoustic conditions, format, provenance and permitted rights. Item-specific licence, gap-fill scope, or both. ### [Your receiving team needs a definition of correct delivery](https://ypai.ai/speech-data/technical-specifications/) - A technical specification for the receiving system: formats, channel and track structure, transcript schema, metadata, validation and the conditions for acceptance. - [Nordic and low-resource languages](https://ypai.ai/speech-data/language-coverage/) - [In-car and automotive voice](https://ypai.ai/voice-recognition-for-automotive/) - [Clinical and healthcare speech](https://ypai.ai/solutions/healthcare/custom-data-collection/) - [TTS voice production](https://ypai.ai/audio/voice-recording/) - Parallel corpora and MTPE, 300+ language pairs - [Ready-made speech datasets](https://ypai.ai/audio/datasets/) A speech type or setting not listed here? YPAI designs the capture protocol around the deployment; describe the system and the condition it runs under. ## Every recording arrives with its consent, metadata and QA trail Four stages. Each adds to the metadata that travels with the audio, so every delivery arrives ready to inspect. ### Coverage specification - The four axes are turned into quotas: which dialects, which settings, which devices, and how much of the edge. Pool depth per language and accent is stated before signature. - **Quotas**: Held within 5 percentage points per cell - **Acceptance**: Criteria written before recording starts ### Capture - Identity-checked speakers record under the specified conditions, with consent per recording. - **Studio**: 48 kHz / 24-bit, SNR 40 dB or better - **Field**: 16 kHz or above, SNR and environment class per file ### Quality gates - Automated technical QC on every file, then native-reviewer transcription checks against the agreed sampling plan. - **Floor**: 98% first-delivery acceptance, re-collected below it - **Evaluation**: WER reported per dialect group ### Delivery - Audio, transcripts and metadata in the formats and location you named, with a signed manifest. - **Pilot**: Output within 10 business days of kick-off - **Destination**: S3, GCS, Azure, SFTP or on-premise Formats, channel and track structure, transcript schema, validation and acceptance are described in full. ## Qualified speakers, not a crowd Every speaker is recruited, identity-checked and qualified for the specific corpus they record in, by dialect region where the corpus needs it. Corpora draw on a network of 210,000+ registered contributors across 50+ countries; each corpus uses only the contributors qualified for it. - **Recruited**: Sourced for the corpus's languages, dialect regions, demographics and settings - **Identity-checked**: Identity on file, consent per recording - **Qualified**: Passes the corpus's own screening recordings before production - **Calibrated**: Transcript agreement tracked per dialect group during production - **Reviewed**: Independent native-linguist review on sampled work ## The consent, rights and residency answers your legal and security review asks for A recording can meet its technical requirements while its permission does not cover the next intended use. Rights, evidence and delivery stay connected, produced as the work runs. What the review asks, and the answer Where a person is in the recording, consent travels with the file. - **Legal basis**: What is the legal basis for each recording?. Consent, per person, per purpose and per session, documented per recording - **Consent**: Can you show consent, per speaker, and what happens on withdrawal?. GDPR Article 7, collected on the YPAI platform; withdrawals quarantined within 72 hours, reported within one business day, erased within 30 days - **Rights**: Which uses does the release cover?. Evaluation, model-training and commercial-deployment permissions defined separately; voice-cloning and resale only by explicit written grant - **Provenance**: Where did each recording come from?. Consent record, capture metadata and SHA-256 hash per file; Croissant 1.1 description and a data card in EU AI Act Article 10 order per dataset - **Residency**: Where does the audio reside, and under whose jurisdiction?. EEA by default, Norwegian jurisdiction; EEA-only sourcing and processing on request - **Agreements**: What governs the engagement on paper?. DPA signed, SCCs available for cross-border transfer ## The vocabulary, the room and the rules change with the industry A clinic, a cabin and a contact centre do not sound alike, and the consent, residency and acceptance rules differ as much as the acoustics. Six industries carry their own speech pages; every other one starts at the intake below. ### [Healthcare](https://ypai.ai/solutions/healthcare/custom-data-collection/) - Clinical speech for ambient documentation and dictation in Nordic languages, recorded under consent frameworks a hospital review can read. - EEA processing by default, consent per recording ### [Automotive](https://ypai.ai/voice-recognition-for-automotive/) - In-cabin voice across dialects, road noise and microphone positions, for wake words, commands and multilingual driver interaction. ### [AI and model companies](https://ypai.ai/solutions/ai-companies/) - Training and evaluation corpora for ASR, TTS and speech-to-speech models, with dialect-balanced sets and results separated per condition. ### [Public sector](https://ypai.ai/solutions/public-sector/) - Citizen-facing speech in Bokmål, Nynorsk and the dialects between them, for service lines, case dictation and accessibility, under Norwegian jurisdiction. - EEA processing by default, Norwegian jurisdiction ### [Industrial and energy](https://ypai.ai/solutions/industrial-energy/) - Voice on the floor and in the field, with hearing protection, machinery noise, radio and far-field capture for inspection and maintenance workflows. ### [Education](https://ypai.ai/solutions/education/) - Learner speech across ages and accents, pronunciation and reading data for language learning and accessibility products. ## From the first brief to a production corpus GDPR Article 7 · EU AI Act Article 10 · DPA included A feasibility read follows inside one EU business day. ### Describe the system - What it has to do, where it fails, and how the resulting data will be used. ### Feasibility read - Languages, dialects, settings and volume checked against pool depth before anything is promised. ### Pilot delivery - Representative output within 10 business days to validate quality gates, formats and the record. ### Production corpus - Weekly deliveries with per-dialect QA, versioning and the signed manifest. ## Consent, rights and audit readiness ### What governance artefacts can be delivered with a speech dataset? Consent records per speaker, capture metadata and a SHA-256 hash per file, demographic and dialect breakdowns, QA audit trails, a Croissant 1.1 description and a data card written in the order EU AI Act Article 10 asks for, and a signed data processing agreement. ### Can the same recordings be used for training, evaluation and a commercial product? Only if the release says so. YPAI defines evaluation, model-training and commercial-deployment permissions separately, with territory, term, withdrawal handling, deletion and retention. If the recordings will feed models you have not named yet, that use is written in rather than assumed later. ### What happens when a speaker withdraws consent? The recordings are quarantined in YPAI systems within 72 hours, the withdrawal is reported to you within one business day with a ledger entry, and erasure completes within 30 days. ### Where is the audio processed and stored? Primary operations are in Norway on EEA infrastructure. Projects can be structured for EEA-only participant sourcing and processing; US or on-premise delivery is arranged where a deployment requires it. ### How is transcription quality measured on dialectal speech? Transcripts are checked by native linguists for the dialect region, agreement is measured per dialect group rather than as one blended average, and disagreements are adjudicated by a dialect-specialist reviewer before the corpus is used for training or evaluation. Bring the system, the corpus or the performance gap. If YPAI is not the right fit, we will say so directly. ## What the platform records for every delivery - consent recorded · 48 kHz · 24-bit · far-field - YPAI's own audio collection platform, with consent recorded per contributor - transcript and speaker turns checked against the guideline - Native human-review console as the platform core - sampled lot · coverage gap · 3 recordings re-collected - Credentialed domain reviewers where the work requires them - manifest · integrity check · erasure within 30 days - Residency, subprocessors and transfers defined per engagement