SilencioNetwork/amharic-speech
收藏Hugging Face2026-04-07 更新2026-04-12 收录
下载链接:
https://hf-mirror.com/datasets/SilencioNetwork/amharic-speech
下载链接
链接失效反馈官方服务:
资源简介:
---
license: cc-by-nc-4.0
language:
- am
task_categories:
- automatic-speech-recognition
- text-to-speech
tags:
- amharic
- ethiopian-languages
- east-africa
- ethiopia
- geez-script
- semitic-languages
- african-languages
- low-resource
- speech-data
- voice-ai
- asr
- tts
pretty_name: "Amharic Speech Dataset"
dataset_info:
features:
- name: file_name
dtype: string
- name: id
dtype: int64
- name: gender
dtype: string
- name: ethnicity
dtype: string
- name: occupation
dtype: string
- name: birth_place
dtype: string
- name: mother_tongue
dtype: string
- name: dialect
dtype: string
- name: year_of_birth
dtype: int64
- name: years_at_birth_place
dtype: int64
- name: languages_data
dtype: string
- name: os
dtype: string
- name: device
dtype: string
- name: browser
dtype: string
- name: duration
dtype: float64
- name: emotions
dtype: string
- name: language
dtype: string
- name: location
dtype: string
- name: noise_sources
dtype: string
- name: script_id
dtype: int64
- name: type_of_script
dtype: string
- name: script
dtype: string
- name: transcript
dtype: string
- name: speaker_id
dtype: string
configs:
- config_name: amharic_ethiopia
data_files:
- split: free_speech
path: amharic_ethiopia/free_speech/**
size_categories:
- n<1K
---
# Amharic Speech Dataset
**The most comprehensive Amharic speech dataset on HuggingFace - natural, real-world Amharic from native speakers in Ethiopia and the diaspora.**
## Dataset Overview
- **Total audio samples**: 51 recordings
- **Total duration**: ~23 minutes
- **Primary region**: Ethiopia (Addis Ababa)
- **Context**: Natural spontaneous speech (free_speech)
- **Audio format**: WAV files
- **Sample rate**: 48 kHz
- **License**: CC BY-NC 4.0 (free for research, non-commercial use)
## Language Context
**Amharic (አማርኛ)** is Ethiopia's primary language:
- **Speakers**: 57M+ (32M native, 25M+ L2)
- **Official language**: Ethiopia (federal working language)
- **Geographic spread**: Ethiopia (primarily central/northern regions)
- **Ge'ez script**: Unique abugida writing system (syllabic alphabet)
- **Linguistic family**: Semitic (Afro-Asiatic) - related to Arabic, Hebrew, Tigrinya
- **Cultural significance**: Ethiopian Orthodox Christianity, Ethiopian literature, music
- **Digital presence**: Growing on social media, YouTube, Ethiopian tech ecosystem
## Target Applications
This dataset is designed for:
- **Amharic ASR systems** - Speech recognition for 57M+ speakers
- **Voice assistants** - Ethiopian tech startups, mobile banking
- **TTS for Amharic** - Text-to-speech with authentic Ethiopian pronunciation
- **Language learning apps** - Pronunciation training for Amharic learners
- **Content moderation** - Social media platforms operating in Ethiopia
- **Transcription services** - Ethiopian media, podcasts, YouTube content
- **Government services** - Voice-enabled public services in Ethiopia
## Dataset Structure
```
amharic-speech/
└── data/
├── audio/ # 51 WAV files
└── metadata.csv # Speaker metadata & transcripts
```
## Data Splits
### Amharic (Ethiopia)
- **Files**: 51 recordings
- **Dialect**: Primarily Addis Ababa (standard Amharic)
- **Context**: Natural spontaneous speech
- **Use case**: General-purpose Amharic ASR, Ethiopian voice AI
## Languages Sampled in This Dataset ✅
51 audio samples available for immediate download:
- **Amharic**: 51 files (~23 minutes)
## Full OTS Inventory Available 📊
This sample represents **<0.19%** of Silencio's complete Amharic speech inventory.
Contact us for access to our full Amharic corpus:
**Amharic by Country:**
- **Ethiopia**: 1,058 hours, 102,378 recordings
- **American Samoa**: 7 hours, 623 recordings
- **Faroe Islands**: 3 hours, 653 recordings
- **Angola**: 2 hours, 311 recordings
- **United States**: 2 hours, 260 recordings
- **Algeria**: 2 hours, 233 recordings
- **Albania**: 2 hours, 83 recordings
- **Honduras**: 1 hour, 234 recordings
- **+ 15 more countries** (diaspora communities)
**Total**: **1,081+ hours** across **105,000+ recordings**
**Contact us for access**: [sofia@silencioai.com](mailto:sofia@silencioai.com)
## Key Features
✅ **Native speakers** - Authentic Ethiopian Amharic (Addis Ababa)
✅ **Natural speech** - Real conversational Amharic, not scripted
✅ **Standard dialect** - Addis Ababa variant (widely understood)
✅ **Diverse topics** - Daily life, opinions, technology, culture
✅ **High audio quality** - 48 kHz WAV format
✅ **Rich metadata** - Gender, dialect, emotions, transcriptions in Ge'ez script
✅ **Ethical data collection** - Consent-based, privacy-preserving
## Use Cases
### 1. Amharic Speech Recognition
Build ASR systems for the 57M+ Amharic-speaking market in Ethiopia.
### 2. Voice Banking & Fintech
Power voice-enabled mobile banking in Ethiopia (M-BIRR, HelloCash, CBE Birr).
### 3. Amharic TTS
Train text-to-speech models with authentic Ethiopian Amharic pronunciation.
### 4. Content Moderation
Build speech detection for Ethiopian social media platforms and YouTube.
### 5. Government Services
Enable voice-based public services in Ethiopia (health, education, agriculture).
### 6. Voice Assistants
Develop Amharic-language voice assistants for Ethiopia's growing smartphone market.
## Loading the Dataset
```python
from datasets import load_dataset
# Load full Amharic dataset
dataset = load_dataset("SilencioNetwork/amharic-speech")
# Access samples
for sample in dataset['train']:
audio = sample['audio']
transcript = sample['transcript']
dialect = sample['dialect']
print(f"Transcript: {transcript}")
print(f"Dialect: {dialect}")
```
## Sample Metadata
Each recording includes:
- `file_name`: Audio file path
- `id`: Unique recording ID
- `gender`: Speaker gender
- `location`: Speaker location
- `mother_tongue`: Native language (Amharic)
- `dialect`: Regional variant (Ethiopia - Addis Ababa)
- `duration`: Recording length (seconds)
- `emotions`: Emotion labels (happy, excited, focused, relaxed, etc.)
- `language`: Amharic
- `type_of_script`: free_speech (spontaneous, unscripted)
- `transcript`: Whisper-generated transcription (Ge'ez script)
- `script`: Original prompt (question asked in Amharic)
## Amharic Speech Characteristics
This dataset captures authentic Amharic speech features:
- **Ge'ez script phonology**: Ejective consonants (ጠ, ቀ, ጨ), labialized consonants
- **Semitic features**: Triconsonantal root system (like Arabic/Hebrew)
- **Complex morphology**: Rich verb conjugation, case marking
- **Tone/stress**: Stress-accent patterns
- **Natural prosody**: Authentic rhythm, intonation
- **Real-world audio**: Mobile recordings, natural environments
## Market Context
### Ethiopian Tech & Economy
- **57M+ Amharic speakers** - Ethiopia's lingua franca
- **Ethiopia**: 120M population (Africa's 2nd most populous), 25M+ internet users
- **Smartphone penetration**: 45% and growing rapidly
- **Digital payments**: Mobile money growing 40%+ annually
- **Tech ecosystem**: Addis Ababa emerging as East African tech hub
- **YouTube**: Ethiopian content exploding (music, news, education)
### Why Amharic Matters
- **Underrepresented in AI**: <0.01% of speech datasets despite 57M+ speakers
- **National language**: Ethiopia's federal working language (government, education, media)
- **Ancient script**: Ge'ez alphabet (one of Africa's oldest writing systems)
- **Growing digital economy**: E-commerce, fintech, edtech booming in Ethiopia
- **Large youth population**: 70% under 30 = massive smartphone adoption potential
## Ge'ez Script
Amharic uses the **Ge'ez script** (also called Ethiopic script):
- **Abugida**: Each character represents a consonant+vowel combination
- **7 vowel orders**: ሀ ሁ ሂ ሃ ሄ ህ ሆ (ha, hu, hi, ha, he, hi, ho)
- **33 base consonants** × 7 vowel orders = 231+ characters
- **Unique to Ethiopia/Eritrea**: Used for Amharic, Tigrinya, Ge'ez (liturgical)
- **Left-to-right**: Unlike Arabic/Hebrew (though same Semitic family)
**ASR/TTS systems need to handle this unique script for written transcription.**
## Ethical Considerations
All data was collected with explicit informed consent from native Amharic speakers. Recordings contain general conversational topics only - no sensitive personal information.
## Comparison to Other Datasets
| Dataset | Language | Hours | Speakers | Natural? |
|---------|----------|-------|----------|----------|
| LibriSpeech | English | 1,000 | 2,484 | ❌ Read speech |
| Common Voice | Amharic | ~20 | Few | ⚠️ Read sentences |
| **Silencio Amharic** | **Amharic** | **1,081+** | **4,500+** | **✅ Spontaneous** |
**This is the largest natural Amharic speech dataset available.**
## Citation
If you use this dataset in your research or commercial product, please cite:
```bibtex
@dataset{silencio_amharic_speech_2026,
title={Amharic Speech Dataset},
author={Silencio Network},
year={2026},
publisher={HuggingFace},
url={https://huggingface.co/datasets/SilencioNetwork/amharic-speech}
}
```
## Related Datasets
- [African Languages Speech](https://huggingface.co/datasets/SilencioNetwork/african-languages-speech) - 6 African languages (Swahili, Hausa, Yoruba, Igbo, Amharic, Nigerian English)
- [Complete Voice AI Speech Dataset](https://huggingface.co/datasets/SilencioNetwork/complete-voiceai-speech-dataset) - 39 language/accent variants
- [Indian Languages Speech](https://huggingface.co/datasets/SilencioNetwork/indian-languages-speech) - 9 Indian languages
- [Yoruba Speech](https://huggingface.co/datasets/SilencioNetwork/yoruba-speech) - 50 Yoruba samples
## License
**CC BY-NC 4.0** (Creative Commons Attribution-NonCommercial 4.0 International)
✅ Free for research and non-commercial use
❌ Commercial use requires licensing (contact us)
## About Silencio
Silencio is a voice AI data sourcing company with 2M+ contributors across 180+ countries. We provide scaled sourcing of real-world audio and speech data for AI labs, robotics companies, and enterprises building voice AI products.
🌐 [silencioai.com](https://www.silencioai.com)
📧 [sofia@silencioai.com](mailto:sofia@silencioai.com)
---
**Tags**: amharic, አማርኛ, ethiopian languages, east africa, ethiopia, geez script, ethiopic script, semitic languages, african languages, low-resource languages, speech recognition, asr, tts, voice ai, natural speech, spontaneous speech, ethiopian speech, addis ababa
---
许可协议:CC BY-NC 4.0
语言:
- 阿姆哈拉语(Amharic)
任务类别:
- 自动语音识别(Automatic Speech Recognition, ASR)
- 文本转语音(Text-to-Speech, TTS)
标签:
- 阿姆哈拉语
- 埃塞俄比亚语言
- 东非语言
- 埃塞俄比亚
- 吉兹字母(Ge'ez Script)
- 闪含语系语言
- 非洲语言
- 低资源语言
- 语音数据
- 语音AI
- ASR
- TTS
友好名称:阿姆哈拉语语音数据集(Amharic Speech Dataset)
数据集信息:
特征:
- 名称:file_name,数据类型:字符串
- 名称:id,数据类型:64位整数
- 名称:gender,数据类型:字符串
- 名称:ethnicity,数据类型:字符串
- 名称:occupation,数据类型:字符串
- 名称:birth_place,数据类型:字符串
- 名称:mother_tongue,数据类型:字符串
- 名称:dialect,数据类型:字符串
- 名称:year_of_birth,数据类型:64位整数
- 名称:years_at_birth_place,数据类型:64位整数
- 名称:languages_data,数据类型:字符串
- 名称:os,数据类型:字符串
- 名称:device,数据类型:字符串
- 名称:browser,数据类型:字符串
- 名称:duration,数据类型:浮点数
- 名称:emotions,数据类型:字符串
- 名称:language,数据类型:字符串
- 名称:location,数据类型:字符串
- 名称:noise_sources,数据类型:字符串
- 名称:script_id,数据类型:64位整数
- 名称:type_of_script,数据类型:字符串
- 名称:script,数据类型:字符串
- 名称:transcript,数据类型:字符串
- 名称:speaker_id,数据类型:字符串
配置项:
- 配置名称:amharic_ethiopia
数据文件:
- 拆分方式:free_speech,路径:amharic_ethiopia/free_speech/**
规模类别:
- 样本数少于1000(n<1K)
---
# 阿姆哈拉语语音数据集
**Hugging Face平台上最全面的阿姆哈拉语语音数据集——收录埃塞俄比亚本土及海外侨民的自然真实阿姆哈拉语口语数据。**
## 数据集概览
- **总音频样本数**:51条录音
- **总时长**:约23分钟
- **主要采集区域**:埃塞俄比亚(亚的斯亚贝巴)
- **采集场景**:自然即兴口语(free_speech)
- **音频格式**:WAV文件
- **采样率**:48 kHz
- **许可协议**:CC BY-NC 4.0(可免费用于科研,非商业用途)
## 语言背景
**阿姆哈拉语(አማርኛ)** 是埃塞俄比亚的官方语言:
- **使用人口**:超5700万(其中母语使用者约3200万,第二语言使用者超2500万)
- **官方地位**:埃塞俄比亚联邦官方工作语言
- **地理分布**:主要集中于埃塞俄比亚中部及北部地区
- **书写系统**:吉兹字母(Ge'ez Script)——独特的音节文字书写体系
- **语系归属**:闪含语系(Afro-Asiatic)闪米特语族,与阿拉伯语、希伯来语、提格利尼亚语同属一类
- **文化意义**:关联埃塞俄比亚东正教、埃塞俄比亚文学与音乐
- **数字化现状**:在社交媒体、YouTube及埃塞俄比亚科技生态中使用率持续增长
## 目标应用场景
本数据集专为以下场景设计:
- **阿姆哈拉语自动语音识别系统**——服务超5700万阿姆哈拉语使用者
- **语音助手**——适配埃塞俄比亚科技初创企业、移动银行业务
- **阿姆哈拉语文本转语音**——生成具有地道埃塞俄比亚口音的语音合成内容
- **语言学习应用**——为阿姆哈拉语学习者提供发音训练素材
- **内容审核**——服务埃塞俄比亚境内的社交媒体平台
- **转录服务**——适配埃塞俄比亚媒体、播客及YouTube内容转录
- **政务服务**——打造埃塞俄比亚境内的语音交互公共服务
## 数据集结构
amharic-speech/
└── data/
├── audio/ # 51条WAV格式音频文件
└── metadata.csv # 包含说话人元数据与转录文本的元数据文件
## 数据拆分
### 阿姆哈拉语(埃塞俄比亚)
- **样本数量**:51条录音
- **方言类型**:以亚的斯亚贝巴标准阿姆哈拉语为主
- **采集场景**:自然即兴口语
- **适用场景**:通用型阿姆哈拉语自动语音识别、埃塞俄比亚语音AI开发
## 本数据集收录的采样语言 ✅
提供51条可立即下载的音频样本:
- **阿姆哈拉语**:51条音频(总时长约23分钟)
## 完整语音库可用 📊
本样本仅占Silencio完整阿姆哈拉语语音库的**<0.19%**。
如需获取完整阿姆哈拉语语料库,请联系我们:
**按国家分类的语料分布:**
- **埃塞俄比亚**:1058小时,102378条录音
- **美属萨摩亚**:7小时,623条录音
- **法罗群岛**:3小时,653条录音
- **安哥拉**:2小时,311条录音
- **美国**:2小时,260条录音
- **阿尔及利亚**:2小时,233条录音
- **阿尔巴尼亚**:2小时,83条录音
- **洪都拉斯**:1小时,234条录音
- **+ 其他15个国家**(海外侨民社区)
**总规模**:**1081小时以上**,覆盖**105000+条录音**
**如需获取授权请联系:** [sofia@silencioai.com](mailto:sofia@silencioai.com)
## 核心特性
✅ **母语使用者录制**——地道埃塞俄比亚阿姆哈拉语(亚的斯亚贝巴口音)
✅ **自然口语**——真实对话场景,非脚本式录制
✅ **标准方言**——亚的斯亚贝巴通用标准口音,受众广泛
✅ **话题多样**——涵盖日常生活、观点表达、科技与文化等主题
✅ **高音质**——48 kHz WAV格式音频
✅ **丰富元数据**——包含性别、方言、情感标签及吉兹字母转录文本
✅ **合规数据采集**——基于知情同意,隐私保护友好
## 应用场景
### 1. 阿姆哈拉语语音识别
为埃塞俄比亚境内超5700万阿姆哈拉语使用者打造自动语音识别系统。
### 2. 语音银行与金融科技
为埃塞俄比亚移动银行业务(如M-BIRR、HelloCash、CBE Birr)提供语音交互能力。
### 3. 阿姆哈拉语文本转语音
训练具备地道埃塞俄比亚阿姆哈拉语发音的文本转语音模型。
### 4. 内容审核
为埃塞俄比亚社交媒体平台及YouTube构建语音检测系统。
### 5. 政务服务
打造埃塞俄比亚境内的语音交互公共服务(医疗、教育、农业领域)。
### 6. 语音助手
为埃塞俄比亚快速增长的智能手机市场开发阿姆哈拉语语音助手。
## 数据集加载代码
python
from datasets import load_dataset
# 加载完整阿姆哈拉语数据集
dataset = load_dataset("SilencioNetwork/amharic-speech")
# 访问样本数据
for sample in dataset['train']:
audio = sample['audio']
transcript = sample['transcript']
dialect = sample['dialect']
print(f"转录文本:{transcript}")
print(f"方言类型:{dialect}")
## 样本元数据说明
每条录音包含以下字段:
- `file_name`:音频文件路径
- `id`:唯一录音ID
- `gender`:说话人性别
- `location`:说话人所在地
- `mother_tongue`:母语(阿姆哈拉语)
- `dialect`:区域方言变体(埃塞俄比亚——亚的斯亚贝巴)
- `duration`:录音时长(单位:秒)
- `emotions`:情感标签(开心、兴奋、专注、放松等)
- `language`:语言(阿姆哈拉语)
- `type_of_script`:脚本类型:free_speech(即兴、非脚本式)
- `transcript`:Whisper生成的吉兹字母转录文本
- `script`:原始阿姆哈拉语提问提示文本
## 阿姆哈拉语语音特征
本数据集收录了地道的阿姆哈拉语语音特征:
- **吉兹字母音系**:挤喉辅音(ጠ、ቀ、ጨ等)、唇化辅音
- **闪含语系特征**:三辅音词根系统(与阿拉伯语、希伯来语类似)
- **复杂形态学**:丰富的动词变位、格标记体系
- **重音与语调**:重音模式与语调特征
- **自然韵律**:真实的节奏与语调变化
- **真实录制环境**:移动设备录制,自然背景噪声场景
## 市场背景
### 埃塞俄比亚科技与经济现状
- **超5700万阿姆哈拉语使用者**——埃塞俄比亚通用语
- **埃塞俄比亚总人口**:1.2亿(非洲第二人口大国),互联网用户超2500万
- **智能手机渗透率**:45%且持续快速增长
- **移动支付**:移动货币业务年增长率超40%
- **科技生态**:亚的斯亚贝巴正逐步成为东非科技枢纽
- **YouTube平台**:埃塞俄比亚本土内容(音乐、新闻、教育类)爆发式增长
### 阿姆哈拉语的重要性
- **AI领域代表性不足**:尽管使用者超5700万,阿姆哈拉语语音数据集占比不足0.01%
- **国家官方语言**:埃塞俄比亚联邦工作语言(覆盖政府、教育、媒体领域)
- **古老书写系统**:吉兹字母(非洲最古老的书写系统之一)
- **数字化经济增长**:埃塞俄比亚电商、金融科技、教育科技行业蓬勃发展
- **年轻化人口结构**:70%人口年龄低于30岁,智能手机 adoption 潜力巨大
## 吉兹字母(Ge'ez Script)
阿姆哈拉语使用**吉兹字母**(又称埃塞俄比亚字母):
- **音节文字体系**:每个字符代表辅音+元音组合
- **7种元音变体**:ሀ ሁ ሂ ሃ ሄ ህ ሆ(对应发音:ha, hu, hi, ha, he, hi, ho)
- **33个基础辅音** × 7种元音变体 = 231+个字符
- **专属区域**:仅埃塞俄比亚与厄立特里亚使用,适配阿姆哈拉语、提格利尼亚语、吉兹语(宗教礼拜用语)
- **书写方向**:从左至右,与阿拉伯语、希伯来语不同(尽管同属闪含语系)
**自动语音识别与文本转语音系统需适配该独特书写系统以完成文字转录。**
## 伦理考量
所有数据均通过阿姆哈拉语母语使用者的明确知情同意完成采集,录音仅包含通用对话主题,不涉及任何敏感个人信息。
## 与其他数据集的对比
| 数据集名称 | 语言 | 总时长 | 说话人数量 | 是否为自然口语 |
|---------|----------|-------|----------|----------|
| LibriSpeech | 英语 | 1000小时 | 2484 | ❌ 朗读式语音 |
| Common Voice | 阿姆哈拉语 | 约20小时 | 较少 | ⚠️ 朗读式句子 |
| **Silencio阿姆哈拉语数据集** | **阿姆哈拉语** | **1081小时以上** | **4500+** | **✅ 即兴口语** |
**本数据集是目前可用的规模最大的自然阿姆哈拉语语音数据集。**
## 引用规范
若您在科研或商业产品中使用本数据集,请引用如下:
bibtex
@dataset{silencio_amharic_speech_2026,
title={Amharic Speech Dataset},
author={Silencio Network},
year={2026},
publisher={HuggingFace},
url={https://huggingface.co/datasets/SilencioNetwork/amharic-speech}
}
## 相关数据集
- [非洲语言语音数据集](https://huggingface.co/datasets/SilencioNetwork/african-languages-speech)——覆盖6种非洲语言(斯瓦希里语、豪萨语、约鲁巴语、伊博语、阿姆哈拉语、尼日利亚英语)
- [完整语音AI语音数据集](https://huggingface.co/datasets/SilencioNetwork/complete-voiceai-speech-dataset)——覆盖39种语言/口音变体
- [印度语言语音数据集](https://huggingface.co/datasets/SilencioNetwork/indian-languages-speech)——覆盖9种印度语言
- [约鲁巴语语音数据集](https://huggingface.co/datasets/SilencioNetwork/yoruba-speech)——51条约鲁巴语语音样本
## 许可协议
**CC BY-NC 4.0**(知识共享署名-非商业性使用4.0国际许可协议)
✅ 可免费用于科研与非商业用途
❌ 商业使用需获得授权(请联系我们)
## 关于Silencio
Silencio是一家语音AI数据采集公司,在全球180+个国家拥有200万+名数据贡献者。我们为AI实验室、机器人公司及语音AI产品开发企业提供真实场景下的规模化语音与音频数据采集服务。
🌐 [silencioai.com](https://www.silencioai.com)
📧 [sofia@silencioai.com](mailto:sofia@silencioai.com)
---
**标签**:阿姆哈拉语、አማርኛ、埃塞俄比亚语言、东非语言、埃塞俄比亚、吉兹字母、埃塞俄比亚字母、闪含语系语言、非洲语言、低资源语言、语音识别、ASR、TTS、语音AI、自然口语、即兴口语、埃塞俄比亚语音、亚的斯亚贝巴
提供机构:
SilencioNetwork


