Datasets:

RobotsMali
/

transcription-scorer

duration float64 0.69 30	text stringlengths 0 463	score float64 0 100
1.1	N'an ka ji min de bɛ yan.	2
2.014	Balɔbɔ ko ayi ne t'i fɛ n ba da la kuma so kɔnɔ na.	50
2.73	N ye so bɛɛ manɔn o kɔrɔ ko mara ma tiɲɛ n bolo.	50
4.31	Ne y'a fɔ mɔgɔ min ye ko ne ye Tunkara, ne ye jeli ye maninkɛ, a b'a fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ ayi fɔ.	35
2.74	À fɔ́, áw bɛ dɔ̀lɔmin ná kɛ́!	60
5.321995	Sàbaminiyan màkari bɛ́ dèli lè lá.	5
2.695063	Mɔgɔmusokɔrɔba bɛ a ɲɔgɔmusokɔrɔba fɛ.	0
1.014	Bɛɛ ye sigira sɔrɔ e la.	30
7.942	Nê man dá yìn kɛ́lɛ ɲɛ́lèli dèli dè, jamà b'í sìgi kà súman, kà súman k'à súman kan ná, nìn sí bɛ́ jɔ̀nkolonin mà.	0
4.323		0
1.892	tɔwari taara de man di, Malikɛn yɛrɛ ye ni kɛlɛ ye	20
8.316		0
2.622	Cɛkɔrɔba kɔnpila t'u la ɲɛdɔn sɔnni kɔnɔ,	0
4.56		0
3.823	I y'a de ye débi de sɔlɔsiyɔn d'o tɛrɛ zɔn ne ye tan, « mais » pɛ p'a régler le problɛme dɛ.	5
11.108	« ne bɛ sisan min na! » Mama ko a fɔra ko aw musomanninw ka kalange kosɛbɛ Yasa ka Baba ni ma bɔn nɔgɔ la y'a mɛ, walayi a fɔ Mama ji tɛ n toro.	50
7.447	Bɛɛ de jijalen don, kosɔbɛ « lakomini » ka sirisi de don na sɛrɛ.	50
1.199	Ka- a-	75
2.286	O kama an ka saliya kɛ i bɛ kan ka kura da i ka yɔrɔ la.	60
4.118	Fan ka sɔrɔ a ma kamalenya, fanka ni kamalenya tɛ kelen.	65
2.789	Nusan Jɔnɔrɔli esɛnsiyɛl dans de nɔbɛrɛse fɔcɔsiɲɔgɔn ye.	0
9.831		0
3.327	An kana fɔ k'an bɛ taa ɲe an bɛ taa ɲe an bɛ dɛsɛ baga an ka denw sanɛ.	60
2.073	Nɔ́n, í b'í ń bɛ̀kiliw fɔ́!	10
2.926	« Maligɛre kɛ san nana i la cɛ. »	75
4.966	Ka waritɛ min yɛrɛ fɔ, ka parisɛnariya jumɛn na, ka nin	50
4.076	I ziniw bɛɛ b'an de fiya « commudo » mɛ o bɛɛ la an « commudo » de ye « piremi » ye.	75
2.413	Zorisɔn kɔntarandi en ouverture de di?	0
7.041	« merci » bɔ sɔrɔ de la.	79
1.988		0
1.81		0
9.87		0
6.737	Unhun!	5
2.13	I bɛ se ka tugu a kɔnɔ na.	0
1.539		0
1.561	K'an ka sigidaw sanni.	0
4.983	Mìmiina ni hàkini ɲuman yé kà ɲɛ̀ sì.	69
1.02		0
1.887	Unhun!	0
2.636	« pour que » vɛrɛman ka sɛgɛsɛgɛliw bɔ.	5
3.003		0
5.617		0
5.685	Lafin bɛ « lacouge » cɛmana sɔn ko min ma?	5
8.972	Ayi awɔ peremanden kɔni « parce que » peremiyɛr ani ansogo fɛn mɛ a ma fɔ ko peremanden ɲe e bɛna erisɛsiyɔn de da wa?	50
4.346	N'u kɔnɔtɛla a ka se o ko bɛ ka ɲinan kosɛbɛ, ee, kuma duman kosɛbɛ k'a miiri bɛ ko bila ka wuli k'a la.	50
2.149	Mìri wó, í ka wó!	50
2.337	N'o ye misiw ye, misiw fɔra farala k'aw n'i b'o dilan komi n'den?	78
2.549		0
5.394	Cɛ tɛrɛ ɲɔpɔrɔtan ka ɲi siso kan « parce que » dɔw kɔnɔ ko Pali a bɛ sɔrɔ.	76
6.104	«An yɛrɛ bɛna an ka ko ɲɛdabɔ an ni ɲɔgɔn cɛ, nka mɔgɔ wɛrɛ sen b'a la. »	55
2.777		0
1.811	Mɛtenan, faamu de bɛ se ka décision ta.	50
7.447	N'aw bɛ kɛlɛn-kɛlɛnni wa	0
2.072	Nêbɛliba kélen kélen nín bɔ́,	5
13.265	Siga foyi tɛ o kuma na, siri bɛ muso jɔfɛn o jɔfɛn bɛ dafɛn ye an ka dugawu dobi ani halusilamɛfure bɛɛ lajɛlen ye.	60
5.984	Ni sinfinna kan nana anw fo ka na laɲiniw kɛ, an kan'u si, an kan'u si, an kan'u kɛ...	70
4.327	Tilenin in bi karidɔgɔn ye jɔn kalo tile tan ni kɔnɔntan.	50
1.123	Sálo nà síni!	50
1.396	Àwɔ!	75
12.498	K'o sabu kɛ bi-bi na o yɛlɛmalen don k'o kɛ banakuntanw sɔrɔyɔrɔ ye, wulu-wulu de bɛ fan bɛɛ, ɲaman bɛ fan bɛɛ, dimɔgɔw bɛ mɔgɔw kan sosow ɲɛmɛrɛ, ɲinɛ, tɛ ee!	55
2.978	Ko caman b'a kɔnɔ, ko caman b'a kɔnɔ.	79
12.875	Na, i mɔgɔsiridimi min ka ɲi hanfɔ dajɛla la ko jamana Siridani.	70
12.728	Sinankuya dabɔ tun tɛ dɔgɔyama ye, nɛni tɛ, waso tɛ, kojugu fɔ tɛ, kɛlɛko tɛ.	70
4.217	N bɛ don miiriw kɔnɔ, mɛ misirilanfɛnama fila bisɛ de b'olu filankolo a kan.	70
7.532	Fulaw bɛ kɛlɛn-kɛlɛn de ye u ɲumakan	70
2.148	Ni bɛɛ y'i ka dalala se dingɛ,	60
10.13	An ka laji an ka sigi k'a fɔcɛ an k'an bolo di ɲɔgɔn ma n'an m'an bolo di ɲɔgɔn ma wa don o don de bɛ kɛ nin ye san o san samiya de bɛ kɛ nin ye.	65
1.069	Kɛsɔn k'u sigira yen.	60
3.31	Anw jɔyɔrɔba bɛ siran, a b'aw lakaran ni ko la Bamakɔ ye « collectivité » ye.	75
11.173	N'o tɛ n'u y'a ko ye sisan k'a fɔ ko n'an yɛlɛla sa yɔrɔw la sa min ka yɔrɔ mara nɔgɔ alamani bɛ bi o kan min ka yɔrɔ mara nɔgɔ a bɛ taa ni tɔgɔ yɔrɔ bɛ cɛ ya.	5
5.175	O de la, an ka jamana kuntigi n'o ye Jinaldi Haramɛ Asimi Koyita ye, n'ale de ye nin jɛkulu ye IESI.	50
1.81	O yɔrɔ tɛ mɔgɔ bɛɛ lajɛlen fo.	40
7.895	Ne ma da in kɛlɛ ɲɛleli, delidenjama b'i sigi ka suma ka suma ka na nin siyɛ Jɔnkolonin.	0
2.11	Alajɔkɔrɔba la.	0
6.326	« en russe de bɔni parange, mɛ e ma fi ne kɛ ka n geto. »	45
1.751	An ka dɔnmakofa ba si tɛ.	5
1.37	Mais o si « parce que » i tɛ se poɔgɔrɔba.	0
3.181	A selen dɛkisɛ la mɔma a fɛkisɛ « manifi tribjɔn en dɔturu numɛ » de ma.	0
2.27	Lɔyɔrɔ don Alokisiyɔn, Razjo Malikɛ.	0
1.803	N'u tɛ muso a kɛ, n'u tɛ fanelen, n'u tɛ banelen.	78
5.818		0
5.473	Fama, ale kɛ ninnu bɔ, an b'a fɔ jumɛn ta?	79
4.439	N'a ye N'Vite a suivre sanze activité la, a nu se den an kɔri a mɛn a senni la.	0
2.07		0
5.78		0
1.143	An y'a ye k’a fɔ	0
1.232	Bi b'i na.	15
7.229	Ne tɔgɔ ye Valeri Dako, ne ye Sanu-Pɛrɛ-Muchɛl tala ɲɛmɔgɔ ye « réseau » parlementaire de faama, minisiri afirikani de siri anw ma.	15
1.826		0
24.592	Unhun ni faama ni magan macuga minɛ.	5
16.639		0
15.381		0
19.074		0
16.535		0
24.096	Aaa a ko bɛ minnu bolo dɔw bɛ dɔw la faamuyali bɛ se ka to a la an bɛ se k'o kɛ an baara taabolo ye ayiwa Jɛnɛbasso ale ni sulimani ɲaari olu wilila ka se mininɛkura kɔnɔna na y'a dɔrɔ aw b'a dɔn k'a fɔ mininɛkura ɲaman ko o ma dogo mɔgɔ si la unun sanga n'a ka waati de don mɛ a bɛ wagati bɔ sisan nɔn fɛmi ye o ɲama in ye o daga bɔra sigida filanan mɔgɔw da la ayiwa nisɔndiya ko don ngɔ!	69
15.427	N'bɛɛ k'i janto	0
17.184	Tiɲɛ yɛrɛ la, ni gudɔrɔn kɛrɛfɛlaw n'olu ma gudɔrɔni ka pavé nu fɛnw k'a la, hali ni y'a furu a tan ke yen yɔrɔn nɔgɔlɔ, ni fiɲɛ kɔni cira, «pucheur » bɛna gudɔrɔn kan diyagoya a bɛ nɔgɔ, sinɔgɔ k'a fɔ ko fɛn wɛrɛ bɛ gudɔrɔn nɔgɔ a ka gɛlɛn parce que le fɛtike mɔbili yɛrɛ bolo a san fɛ, mɔbili fiɲɛ yɛ	76
22.225	Numɔlɔ min bɛ jurufɔ, numɔnɛ ye, numɔnɛ de kisiw tɛ kabana juru, u bɛ n'bolo mɛ jurufɔw ye kalaba.	70
21.247		0
27.873	Dɔnsɔn, cɛntɔrɔba sɔn, fɔ la tɛ, mɔnɔntidon si tɔgɔlen digɛ.	0

End of preview. Expand in Data Studio

Transcription Scorer Dataset

The Transcription Scorer dataset was created to support research in reference-free evaluation of Automatic Speech Recognition (ASR) systems using human feedback. Unlike traditional evaluation metrics such as WER and its derivatives, this dataset reflects judgments of ASR outputs by human raters across multiple criteria, simulating the way a teacher grades students.

⚙️ What’s Inside

This dataset contains 1200 audio samples (from diverse sources including music with lyrics) totaling 2.28 hours. It is made of short to meduim length segments each associated with:

One transcriptions (drawn by selecting the best hypothesis of two Bambara ASR models)
A score between 0 and 100 assigned by human annotators

bucket (s)	partially‑reviewed
0.6 – 15	965
15 – 30	235

Sources:

Transcriptions were generated by two ASR models:
- Djelia-V1 (proprietary, access through API)
- Soloni (open-source from RobotsMali)
Additional 81 transcriptions were intentionally randomized/shuffled to measure baseline judgment.

Most of the audios were collected by RobotsMali AI4D Lab with the Office de Radio et Télévision du Mali which gave us early access to a few archives of some of their past emissions in Bamanankan. But this dataset also include a few samples from bam-asr-early.

The evaluation was based on the following criteria but we also left room for a personnal subjective judgement so it also include some form of human preference feedback as the annotations were partially reviwed by professional Bambara linguists. So it is a Human feedback dataset but not based on preferences only, the score is actually designed to be a refective of the quality of the transcriptions enough to be used as a proxy metric.

Usage

This dataset is intended for researchers and developers who face a label scarcity situation making traditional ASR evaluation metrics like WER impossible (which is especially relevent to low resource languges such as Bambara). By leveraging human-assigned scores, it enables the development of scoring models which outputs can be used as a proxy to transcription quality. Whether you're building evaluation tools or studying human feedback in speech systems, you might find this dataset useful if you face label scarcity.

Developing reference-free evaluation metrics
Training reward models for RLHF-based fine-tuning of ASR systems
Understanding how human preferences relate to transcription quality

from datasets import load_dataset

# Load the dataset into Hugging Face Dataset object
dataset = load_dataset("RobotsMali/transcription-scorer", "partially-reviewed")

Data Splits

Train: 1000 examples (~1.92h)
Test: 200 examples (~0.37h)

This initial version is only partially reviewed, so you may contribute by opening a PR or a discussion if you find that some assigned scores are innacurate.

Fields

audio: raw audio
duration: audio length (seconds)
transcription: text output to be scored
score: human-assigned score (0–100)

Known Limitations / Issues

Human scoring may contain inconsistencies.
Only partial review/consensus exists — scores may be refined in future versions.
The dataset is very limited in context diversity and transcription variance, only two models were used to generate transcriptions for the same ~560 audios + 80 shuffled transcriptions for baseline estimation so it will benefit from additional data from different distribution.

🤝 Contribute

Feel something was misjudged? Want to improve score consistency? Add transcriptions from another model ? Please open a discussion — we welcome feedback and collaboration.

📜 Citation

@misc{transcription_scorer_2025,
  title={A Dataset of human evaluations of Automatic Speech Recognition for low Resource Bambara language},
  author={RobotsMali AI4D Lab},
  year={2025},
  publisher={Hugging Face}
}

Downloads last month: 56

Size of downloaded dataset files:

696 MB

Size of the auto-converted Parquet files:

696 MB

Number of rows:

1,200