Speech & audio data · ready-made dataset

Pakistani telephony corpus

Telephony is its own acoustic domain. An 8 kHz narrowband channel discards most of the spectral information a wideband model learned from, which is why studio-trained ASR falls apart on real calls. This corpus is collected on the channel your model will actually be deployed to.

Get a free sample Ask about a subset or pilot

Free unwatermarked samplePer-record consentTraining-rights licence14-day acceptance

Datasheet summary

What is in Pakistani Telephony Corpus

8 kHz narrowband call-center audio, channel-separated agent/caller stems, GSM and VoIP codecs, intent and outcome labels.

Dataset ID	TDA-SP-002
Modality	Speech
Languages / coverage	Urdu, Punjabi, Sindhi, Pashto
Typical first delivery	200–500 hrs
Free sample	2–5 hours of audio with transcripts and full metadata, within one business day
Provenance	Collected in Pakistan; collection date, location, contributor pseudonym, device profile and consent version on every record
Licence	Perpetual, worldwide, sublicensable; derivative-model rights stated explicitly. Exclusivity optional. Licensing detail
Delivery	Your S3, GCS or Azure bucket with manifests, checksums, QA report and datasheet
Guarantee	14-day acceptance window; anything outside signed tolerance re-collected free

Included in every delivery

8 kHz narrowband call-center audio
Channel-separated agent and caller stems
GSM and VoIP codec conditions
Intent and outcome labels per call
Urdu, Punjabi, Sindhi and Pashto

Built for

Call-center ASR and agent-assist for Pakistani-language contact centres
IVR and voice-bot intent classification on narrowband audio
Codec-robustness training and evaluation (GSM and VoIP)
Conversation outcome prediction and QA scoring

Evaluate before you buy

Start with the free sample, then a pilot

Request 2–5 hours of audio with transcripts and full metadata from Pakistani Telephony Corpus — identical in format and quality to the paid corpus, no watermark, no sales call. Pilots are creditable against production and start within five business days of a signed SOW.

Request the sample pack Scope a pilot

Average first response under 6 business hours. NDA signed same day.

Consent form, licence text and DPA available before you commit
Datasheet lists composition, QA method and known limitations
Subsets by language, region or condition on request
Custom collection to your spec using the same protocol

Answers

Questions about this dataset

Downsampling removes bandwidth but not the codec artefacts, packet loss, handset variation and background conditions of real calls. Models trained on downsampled studio speech still degrade sharply on production telephony.

Yes. Stems are channel-separated so you can train and evaluate on either side independently, or on the mixed signal.

Each call carries intent and outcome labels alongside the transcript. The label schema is documented in the datasheet that ships with every delivery and with the free sample.

Related datasets

Also in the catalog

Speech