SaarAI/GSMAEthioTelecomBench

GSMA EthioTelecomBench: Amharic ASR Benchmark for Telecom Domain

Overview

GSMA EthioTelecomBench is a comprehensive benchmark for evaluating Automatic Speech Recognition (ASR) systems on Amharic speech, with a focus on telecom customer service conversations. This dataset contains evaluation results from 12 models across 7 evaluation splits.

Audio Conditions

The benchmark evaluates models across different audio conditions:

Column Name	Description
Clean Studio	High-quality studio recordings
Ambient Room Noise	Standard recordings with background noise
Harsh Environment	Heavily degraded audio conditions
Distant / Low-Volume Mic	Low-quality microphone recordings
Wideband Phone (VoLTE)	Real telecom VoLTE recordings
Randomized Phone Channel	Random phone channel conditions
Narrowband 2G + Packet Loss	Degraded 2G network with packet loss

Text Normalization

All WER and CER metrics are computed after applying full text normalization to both reference and predicted transcriptions:

1. Number-to-Text Conversion: Arabic numerals are converted to Amharic text (e.g., "123" → "አንድ መቶ ሃያ ሶስት")
2. Punctuation Removal: All punctuation marks (including Ge'ez punctuation ፠፡።፣፤፥፦፧፨) are removed
3. Character Normalization: Ge'ez character variants are normalized to canonical forms:
   • ሀ/ሐ/ኅ/ኻ/ኃ → ሃ
   • ሠ/ሡ/ሢ/ሣ/ሤ/ሥ/ሦ → ሰ/ሱ/ሲ/ሳ/ሴ/ስ/ሶ
   • ዐ/ዑ/ዒ/ዓ/ዔ/ዕ/ዖ → አ/ኡ/ኢ/አ/ኤ/እ/ኦ
   • ፀ/ፁ/ፂ/ፃ/ፄ/ፅ/ፆ → ጸ/ጹ/ጺ/ጻ/ጼ/ጽ/ጾ
   • And other labialized character normalizations
4. Whitespace Normalization: Multiple spaces collapsed to single space

Evaluated Models

The benchmark includes the following model families:

Ethio-ASR Models (badrex)

• badrex/Ethio-ASR-amharic - Amharic-specific model
• badrex/Ethio-ASR-multilingual-94M - 94M parameter multilingual
• badrex/Ethio-ASR-multilingual-300M - 300M parameter multilingual
• badrex/Ethio-ASR-multilingual-600M - 600M parameter multilingual

OmniASR Models

• omniASR_CTC_300M_v2 - CTC-based 300M
• omniASR_CTC_1B_v2 - CTC-based 1B
• omniASR_CTC_3B_v2 - CTC-based 3B
• omniASR_LLM_300M_v2 - LLM-based 300M
• omniASR_LLM_1B_v2 - LLM-based 1B
• omniASR_LLM_3B_v2 - LLM-based 3B

Baseline Models

• facebook/mms-1b-all - Meta's Massively Multilingual Speech
• openai/whisper-small - OpenAI Whisper Small

Dataset Files

This dataset contains three CSV tables:

1. WER.csv

Word Error Rate (WER) scores for all models across all audio conditions. Lower is better.

2. CER.csv

Character Error Rate (CER) scores for all models across all audio conditions. Lower is better.

3. Gender_GAP.csv

Gender fairness analysis showing:

• Male WER/CER scores
• Female WER/CER scores
• Gender gap (Female - Male): Positive values indicate higher error rates for female speakers

Key Findings

Gender Fairness Analysis

Average WER gender gap (Female - Male) across models shows female speakers are consistently disadvantaged across all audio conditions, with gaps ranging from +3.77% to +13.64%.

Key Observations

1. Domain Adaptation Matters: Models fine-tuned on Amharic (Ethio-ASR family) significantly outperform general multilingual models on telecom domain data.

2. Gender Bias: Most models show higher error rates for female speakers, indicating a systematic gender bias in ASR performance.

3. Audio Quality Impact: Performance degrades significantly on low-quality and over-augmented audio, highlighting the need for robust ASR systems.

4. Model Size vs Performance: Larger models (3B parameters) generally perform better, but domain-specific smaller models can be competitive.

Usage

from datasets import load_dataset

# Load WER scores
wer_data = load_dataset("SAARAI/GSMAEthioTelecomBench", data_files="data/WER.csv", split="train")

# Load CER scores
cer_data = load_dataset("SAARAI/GSMAEthioTelecomBench", data_files="data/CER.csv", split="train")

# Load Gender Gap analysis
gender_data = load_dataset("SAARAI/GSMAEthioTelecomBench", data_files="data/Gender_GAP.csv", split="train")