Echo voices and Baobab national models — Africa-first language intelligence.

Explore Baobab
Skip to content
Open consoleLog in

National model card

LugemiBaobab Twi

Structured status — not a raw JSON dump.

National Twi speech + MT corridor (phrase + optional NC NLLB) with Phase D generative research slot — generative_inference_live false; trainingEligible locked.

Translate (en → Twi)Baobab canopyBaobab GaNationalUnlockSLOResearchAll status cardsRaw API JSON

Overview

Identity & capabilities

Model

deferred_slice_shipped
model_id
baobab-twi
display_name
Baobab Twi
model_version
2026.10.deferred
family
baobab
status
deferred_slice_shipped
as_of
2026-10-23T09:41:36.282Z
product_line
National Twi speech + MT corridor (phrase + optional NC NLLB) with Phase D generative research slot — generative_inference_live false; trainingEligible locked.

capabilities

phrase_mt
phrase_table_only
phrase_mt_phase_b_domains
true
national_api_tiers
true
generative_lm_research_slot
true
generative_inference_live
false
neural_nmt
false
neural_asr
false
neural_tts
false
s2s_corridor
true
foundation_llm_weights
false

honesty flags

Flags are published as returned by the API — unmet gates stay false.

trainingEligiblefalse
neural_nmtfalse
neural_speech_livefalse
generative_inference_livefalse
foundation_llm_weightsfalse
ships_trained_baobab_foundationfalse
phrase_table_onlytrue
asante_akuapem_interchangeablefalse
variety_exclusive_speech_finetunesfalse
bare_ak_defaults_to
ak-gh-asante
bare_ak_warning
Bare `ak` / `tw` / `twi` resolves to Asante (`ak-gh-asante`). Pass `ak-gh-akuapem` or `ak-akuapem` for Akuapem. Varieties are never interchangeable. Shared Akan checkpoints + orthography — not exclusive finetunes.
license_warnings
  • MMS TTS (mms-tts-aka) is CC-BY-NC-4.0 — unrestricted commercial redistribution of those weights is blocked.
  • Whisper-twi-v2 is Apache-2.0; attribute and pin versions.
  • NLLB-200 distilled (baobab-twi-mt) is CC-BY-NC — commercial_license_cleared false; fail-closed phrase MT remains available.
  • Phrase/lexical Baobab MT covers reviewed domains; neural NMT when live is NC-restricted.

varieties

Asante and Akuapem packs

Separate corridors — never interchangeable. ASR / TTS / MT status is per variety.

asante

Asante Twi

ak-gh-asante
bcp47
ak-GH
corridor_id
en-ak-asante
primary_use
Everyday / CS / MeetingBridge Ghana
language_pack_version
asante-pack-2026.10-b
echo_voice_id
own:ak-gh-female
english_accent
own:en-gh-female · en-GH · Ama · Ghanaian English
mt_lang_key
ak
trainingEligible
false
domains
  • greetings
  • public_services
  • education
  • health
  • civic
  • customer_support
not_interchangeable_with
  • ak-gh-akuapem
  • ak-gh-akuapem-twi
  • ak-gh-fante

mt

status
phrase_table_only
pairs
en↔ak

Phase B phrase + lexical pack. Fail-closed when uncovered. Neural NMT when BAOBAB_TWI_NMT_URL live.

asr

status
neural_offline
checkpoint
tiffany101/whisper-twi-v2
license
Apache-2.0
speech_stack
Asante speech
variety_exclusive_finetune
false

tts

status
neural_offline
checkpoint
onnx-community/mms-tts-aka-ONNX
license
CC-BY-NC-4.0
speech_stack
Asante speech
variety_exclusive_finetune
false

s2s_corridor

status
corridor_pass

Corridor scripts PASS (en-GH TTS ↔ Baobab ↔ Asante TTS/ASR). Broad domain deferred.

akuapem

Akuapem Twi

ak-gh-akuapem
bcp47
ak-GH
corridor_id
en-ak-akuapem
primary_use
Education, literary, broadcast orthography
language_pack_version
akuapem-pack-2026.10-b
echo_voice_id
own:ak-gh-akuapem-female
english_accent
own:en-gh-female · en-GH · Ama · Ghanaian English
mt_lang_key
ak-akuapem
trainingEligible
false
domains
  • greetings
  • public_services
  • education
  • health
  • civic
  • customer_support
not_interchangeable_with
  • ak-gh-asante
  • ak-gh-asante-twi
  • ak-gh-fante

mt

status
phrase_table_only
pairs
en↔ak-akuapem

Separate Akuapem Phase B phrase table. Fail-closed when uncovered.

asr

status
neural_offline
checkpoint
tiffany101/whisper-twi-v2
license
Apache-2.0
speech_stack
Akuapem / Asante speech
variety_exclusive_finetune
false

tts

status
neural_offline
checkpoint
onnx-community/mms-tts-aka-ONNX
license
CC-BY-NC-4.0
speech_stack
Akuapem / Asante speech
variety_exclusive_finetune
false

s2s_corridor

status
corridor_pass

Corridor scripts PASS. Upstream speech is Akan/Twi-general; variety via orthography + rescoring.

national_tiers

Who can call what

national_tiers

Public demo

public_demo
who
Citizens, press, students (unauthenticated demos)
auth
None — soft abuse controls on demo speech
maps_to_plan
null
audit
false
commercial_nc_clean_tts_required
false
endpoints
  • GET /v1/models/baobab-twi
  • GET /v1/baobab/twi/national
  • POST /v1/demo/speech
controls
  • honest en-GH remap when speech URL offline (paired Ghanaian English accent)
  • variety headers on Akan demos
limits
translate_chars_daily_hint

Authenticated translate required for production MT

demo_speech

400 requests / 10 minutes / IP (demo controller)

rate_limit_note

Demo speech returns rate_limited when exceeded

national_tiers

Education / free

education
who
Students, KNUST labs, research orgs on Free/Pro plans
auth
Org API key (TranslateAuthGuard) — Free or Pro plan
maps_to_plan
free|pro
audit
true
commercial_nc_clean_tts_required
false
residency_preference
GH
endpoints
  • POST /v1/translate
  • POST /v1/tts/synthesize
  • POST /v1/audio/transcriptions (when ASR keyed)
  • GET /v1/models/baobab-twi
controls
  • UsageService metering
  • BillingService.assertWithinQuota
limits
translate_chars_daily_hint

Free: BILLING_FREE_TRANSLATE_CHARS_QUOTA (default 50k); Pro: BILLING_PRO_TRANSLATE_CHARS_QUOTA

rate_limit_note

RATE_LIMIT_FREE_PER_KEY / RATE_LIMIT_PRO_PER_KEY via existing RateLimitGuard

national_tiers

Ministry production

ministry
who
Ghana government services
auth
Ministry-keyed org API key — Business plan mapping
maps_to_plan
business
audit
true
commercial_nc_clean_tts_required
false
residency_preference
GH
endpoints
  • POST /v1/translate
  • POST /v1/tts/synthesize
  • MeetingBridge Ghana corridor
  • GET /v1/residency
  • GET /v1/metrics/translate
controls
  • AuditService on translate.completed
  • Org retention / vendor-training settings
  • Residency preference GH (affinity, not hard geo-fence)
limits
translate_chars_daily_hint

Business translate quota (BILLING_BUSINESS_TRANSLATE_CHARS_QUOTA)

rate_limit_note

RATE_LIMIT_BUSINESS_PER_KEY / PER_ORG

national_tiers

Enterprise

enterprise
who
Banks, telcos, NGOs
auth
Paid Enterprise keys
maps_to_plan
enterprise
audit
true
commercial_nc_clean_tts_required
true
residency_preference
AF
endpoints
  • POST /v1/translate (+ TM / quality)
  • Speech + MeetingBridge
  • Fine-tunes / marketplace where plan allows
controls
  • SLA targets (documented; contractual SLA separate)
  • Commercial TTS path must be NC-clean for redistribution of MMS weights
  • Audit + retention
limits
translate_chars_daily_hint

Enterprise translate quota

rate_limit_note

RATE_LIMIT_ENTERPRISE_PER_KEY / PER_ORG

research · acceptance_gates

Generative LM research slot

Research slot + harness + training recipe only. Do not brand phrase/NLLB MT or speech corridors as a full national LLM. Generative inference unsupported until gates are met and Lugemi-owned weights exist.

baobab-twi-generative-lm

research_planned
generative_inference_livefalse
foundation_llm_weightsfalse
ships_trained_baobab_foundationfalse
separate_variety_checkpoints_requiredtrue
proposed_checkpoint_ids.asante
baobab-twi-asante-lm-*
proposed_checkpoint_ids.akuapem
baobab-twi-akuapem-lm-*
proposed_checkpoint_ids.never_one_blob
true

acceptance_gates

  • held_out_speakersrequired truemet false

    Held-out evaluation speakers must never appear in training exports; speaker lists remain empty until consented collection.

  • community_consent_asante_and_akuapemrequired truemet false

    trainingEligible stays false until community Asante and Akuapem speakers re-confirm for training.

  • separate_variety_adaptersrequired truemet false

    Any future LM/adapters must ship as baobab-twi-asante-* and baobab-twi-akuapem-* — never one blob marketed as both.

  • beat_phrase_mt_baselinesrequired truemet false

    Generative claims require held-out gains vs corridor phrase-MT baselines on declared Twi-native tasks (Mix/Fidelity/Live gates).

  • commercial_rights_clearrequired truemet false

    Base model + data licenses must clear commercial ministry/enterprise redistribution before promotion.

training_eligible_unlock

trainingEligiblefalse
statuslocked

trainingEligible remains false until all gates are met with durable evidence. Do not invent consent.

  • community_consent_asanterequired truemet false

    Asante community speakers must reconfirm training export consent.

    evidence_required Signed Asante contributor agreements reconfirming training use

  • community_consent_akuapemrequired truemet false

    Akuapem community speakers must reconfirm training export consent.

    evidence_required Signed Akuapem contributor agreements reconfirming training use

  • held_out_speaker_listsrequired truemet false

    Held-out evaluation speakers must never appear in training exports.

    evidence_required Published held-out speaker IDs never present in train splits

  • dual_variety_native_reviewrequired truemet false

    Native review must cover both varieties separately before any train export.

    evidence_required Paid Asante + Akuapem native review records with trainingEligible fields

  • rights_clear_for_exportrequired truemet false

    IP / commercial rights must clear before trainingEligible flips.

    evidence_required Legal memo clearing commercial vs research split for export corpus

honesty.residency

Serving & Accra affinity

serving_today
jnb
serving_today_label
Africa-primary Fly region Johannesburg (jnb)
affinity_target
af-west-1
affinity_city
Accra · GH
accra_gpu_island
false
accra_gpu_island_status
deferred
hard_geo_fence
false
preference_model
affinity_not_hard_fence
affinity_enforcement
soft_preference_recorded
note
Inference prefers matching person/org residency to nearby model hosting when ready — affinity, not a hard geo-fence. Accra (af-west-1) is the Ghana affinity target; live serving for verbalab + lugemi-asante + baobab-twi-mt is jnb today. No Accra GPU island is deployed. Ministry requests record X-Lugemi-Residency-Affinity soft pin.

endpoints · docs

API surface & references

endpoints

model_card
GET /v1/models/baobab-twi
national
GET /v1/baobab/twi/national
research
GET /v1/baobab/twi/research
unlock
GET /v1/baobab/twi/unlock
slo
GET /v1/baobab/twi/slo
translate
POST /v1/translate
demo_speech
POST /v1/demo/speech
tts
POST /v1/tts/synthesize
meetingbridge_ghana
MeetingBridge Ghana corridor
dataset_manifests
apps/api/eval/baobab-twi/dataset-manifest.*.json
residency
GET /v1/residency
translate_metrics
GET /v1/metrics/translate

Machine-readable card: https://api.lugemi.com/v1/models/baobab-twi

docs

national_model
docs/baobab/TWI_NATIONAL_MODEL.md
model_card
docs/baobab/baobab-twi.MODEL_CARD.md
partnership_open_eval
docs/baobab/PARTNERSHIP_OPEN_EVAL.md
ministry_sla
docs/baobab/MINISTRY_SLA.md
training_eligible_unlock
docs/baobab/TRAINING_ELIGIBLE_UNLOCK.md
speech_finetune_recipe
docs/baobab/SPEECH_FINETUNE_RECIPE.md
generative_lm_recipe
docs/baobab/GENERATIVE_LM_RECIPE.md
accra_residency
docs/baobab/ACCRA_RESIDENCY.md
asante_corridor
en-ak-asante
akuapem_corridor
en-ak-akuapem
language_kit
docs/models/05_LANGUAGE_KIT.md
residency
docs/residency.md

Try the corridor.

Run Translate, open the Baobab canopy, or fetch the same fields as JSON for clients.

TranslateBaobabModelsRaw JSON