Ethiopia + China · Data delivery
Vision annotation in production.
Ethiopian speech collection
as the next line.
Shanghai Qiming Instruments runs a 30-person delivery floor in Addis Ababa through our Ethiopian entity, Africa Star Human Resources. Six years of industrial computer-vision work for automotive, surveillance and optics clients. From 2026, speech corpora in six Ethiopian languages — plus Mandarin and Cantonese through our Shanghai entity. Two sites, one quality process.
Lines of work
One team, two data types
Vision is the line we have delivered at volume for six years. Speech is what we are building in 2026 with the same floor, the same supervisors and the same double-check before anything ships.
Computer vision annotation
Industrial inspection, automotive and imaging datasets. Delivered to client schemas, not ours.
- 2D annotation — bounding box, polygon, keypoint
- 3D point cloud and LiDAR labelling
- Semantic and instance segmentation
- Attribute and scene classification
- OCR and document structuring
- Double QC — Addis self-check, then Shanghai engineer spot-check before delivery
Speech and multilingual corpus
Collected in-house by native speakers — not brokered through a crowd platform.
- Six Ethiopian languages, recorded by our own staff
- Mandarin and Cantonese via our Shanghai entity and vetted subcontractors
- Two-speaker natural dialogue, scripted reading, call-centre audio, domain dialogue
- Domains: agriculture, finance, telecom, health
- 16-bit / 48 kHz WAV, speaker-split channels, time-aligned transcripts
- 30 seats live today; 100 within two quarters, recruited through the same university channel
am-ETzh-CNom-ETyue-CNti-ETso-ETwal-ETgez-ETWhy Addis Ababa
What the site actually solves
Every collection market has one constraint that decides whether a project ships on time. Here it is grid power — and it is the reason we built the floor the way we did.
- Full-time staff, not a crowd pool All 30 seats are permanent employees, university-educated and taught in English. They read your guideline directly, which removes the drift that appears when instructions pass through a broker.
- Addis Ababa University channel New speaker cohorts are recruited through university networks, with age, gender, region and dialect quotas applied at intake rather than after the fact. Roughly two weeks from instruction to a screened cohort.
- Power redundancy, sized for recording On-site diesel generator plus UPS, configured so a session does not stop when the grid drops.
- Workstations rated for the work Every seat runs an RTX-class GPU. Point-cloud and LiDAR batches are worked and reviewed locally, not queued behind a remote machine.
The constraint
Average for firms operating here
It is the single most common reason recording projects fail in Ethiopia — not recruitment, not language coverage, not consent paperwork.
Sessions get cut mid-take, a speaker's hour is lost, and the schedule slips quietly until the deadline. We built around it from the start: generator and UPS on the recording floor, and a session plan that assumes an interruption will happen. That is why we can commit to a delivery date rather than a delivery estimate.
Specification
Speech delivery spec
These are our defaults. Where you already have a schema, an orthography convention or a metadata dictionary, we work to yours and document the mapping once at kickoff.
| Audio format | 16-bit / 48 kHz WAV Other sample rates and codecs on request |
|---|---|
| Channels | Two channels, one per speaker Single mixed track also available |
| Transcription | Time-aligned verbatim transcript, per speaker Ge'ez script for Amharic and Tigrinya; client convention elsewhere |
| Speaker metadata | Age band, gender, region, dialect, first language, recording environment |
| Recruitment | Quota-based on age, gender, region and dialect Screening call recorded and retained against each speaker ID |
| Rights | Signed commercial-use consent for every speaker Consent form filed against the speaker ID, retrievable together with the audio |
| Environment | Quiet studio on the delivery floor Field recording for ambient realism where the domain calls for it |
| Pilot scope | 20–50 hours, delivered against agreed acceptance criteria |
| Tooling | CVAT and Label Studio in-house We work inside your platform where you already have one |
Samples
Three stations on the line
Screens from vision systems running on customer equipment. Work of this type is what trained the QC process we now apply to audio.
Clients and compliance
Who we work for, and how the data is held
Client names are withheld under confidentiality. The sectors and programme types below are accurate and can be discussed in more detail under NDA.
Programme history
- Chinese electric vehicle manufacturerMulti-year industrial vision annotation programme
- Global automotive Tier 1 supplierManufacturing inspection datasets
- Global video-surveillance manufacturerDetection and classification datasets
- Global imaging and optics companyImage quality and scene annotation
Data rights
- Ethiopia's Personal Data Protection Proclamation No. 1321/2024 governs collection on Ethiopian soil. Our collection protocol is written against it, including how a contributor withdraws.
- Signed consent before recording. Every contributor signs a commercial-use authorisation; the form is filed against the speaker ID and handed over with the corpus.
- Project-level isolation. Separate storage and a named access list for each engagement. Audio from one client is never visible to another team.
- NDA before samples. We can sign yours, or send ours, before any sample audio or annotation export leaves the building.
Cross-border transfer terms are set in the contract and reviewed against Proclamation 1321/2024 for each engagement. Tell us your jurisdiction at the scoping stage and we will confirm what applies.
Procurement questions
Questions buyers ask first
Straight answers, including the one where the answer is no.
What is the minimum pilot batch?
20 hours of audio is the floor for a speech pilot; 50 hours is the practical size if you want domain spread across several speakers and conditions. For vision, the minimum is a single batch defined by your acceptance criteria — we would rather scope it with you than quote a fixed number that fits nobody.
Who writes the acceptance criteria?
You do. We take your criteria, turn them into a written annotation or recording guideline with worked examples, and send it back for sign-off before any production starts. If you do not have criteria yet, we draft a set from your intended use and you approve or change it. Nothing is delivered against an unstated standard.
How is rework handled?
Work that misses the agreed criteria because of our error is redone at our cost and does not consume the batch. If the criteria change after sign-off, we re-scope and re-price the affected portion rather than absorbing it silently. Disputes are settled against the signed guideline, which is why we insist on one.
Can the data leave Ethiopia, and how?
Yes, under a written transfer arrangement. Ethiopia's Proclamation 1321/2024 is the governing text; the applicable terms depend on your jurisdiction and on whether the data is personal data in the Ethiopian sense. Encrypted handover, retention period and deletion are all fixed in the contract, and consent forms travel with the corpus so your own compliance review has the paper trail.
Who runs a Chinese-language project?
Our Shanghai entity, working with vetted subcontractors in mainland China. Shanghai also owns the QC function for every project, so Mandarin and Cantonese delivery goes through the same review gate as Ethiopian audio — just without the second site.
What error rate should we expect?
We will not quote one before measuring. Your pilot is the measurement: we deliver the batch, you score it against your criteria, and the observed rate becomes the baseline written into the production contract. Teams that quote a percentage up front are usually quoting a different dataset.
Payment terms and currency?
Invoiced in USD by the Shanghai entity, by bank transfer. The pilot is invoiced separately and paid on delivery, so neither side carries a long commitment before the work is proven. Production terms are agreed in the contract and can be milestone-based.
Do you hold ISO 27001 or ISO 17100?
No. We are not certified, and we are not going to imply otherwise. Certification is in progress rather than complete; until then we work to a written collection and access-control policy that we will send you in full, and we accept client security review as part of onboarding.
Start with a pilot
A 20–50 hour batch against your acceptance criteria settles the question faster than any capability deck. Send the languages, the domain and the format you need; we reply with a scoped plan, a timeline and a price for the batch.
Book a 20-minute call- Contact
- Mr. Liu, Managing Director
- liuxin@qimingdata.cn
- Telephone
- +86 177 0805 3706 · Shanghai
+251 97 774 3423 · Addis Ababa - Shanghai
- No. 325-A3 Dakang Road, Baoshan District, Shanghai 200072, China
- Addis Ababa
- Africa Star Human Resources
Jonny Garage, Addis Ababa, Ethiopia