Add README.md

README.md

@@ -16,7 +16,7 @@ tags:
   - speech-to-text
 pretty_name: KenSpeech
 size_categories:
-  - 10K<n<100K
+  - 1K<n<10K
 ---
 
 # KenSpeech: A Swahili Speech Dataset for ASR
@@ -35,209 +35,154 @@ size_categories:
 | Total Speakers | 26 |
 | Female Speakers | 19 |
 | Male Speakers | 7 |
-| Total Transcript Files | 7,939 |
 | Lexicon Words | 31,728+ |
 
 ## Audio Format
 
-All audio recordings are standardized to the following format:
-
 | Property | Value |
 |----------|-------|
-| Format | WAV |
-| Bit Depth | 16-bit |
+| Format | WAV/MP3/MP4/M4A |
 | Sample Rate | 16 kHz |
 | Channels | Mono |
-| Byte Order | Little Endian |
 
 ---
 
-## Dataset Components
-
-### 1. Speech Transcripts (`stt_transcripts/`)
-
-Contains **7,939 transcript files** with corresponding audio transcriptions. Each transcript file is named to match its corresponding audio file.
-
-**Example:**
-- Audio: `tweet_5701.wav`
-- Transcript: `tweet_5701.txt`
-
-### 2. Dictionary Dataset (`stt_dictionary/`)
-
-Contains audio recordings organized by speaker, along with corresponding transcripts, metadata, and a pronunciation lexicon.
+## Dataset Structure
 
-#### Structure:
 ```
-stt_dictionary/
-├── metadata.csv              # Main metadata file linking audio to transcripts
-├── lexicon.csv               # Pronunciation dictionary (31K+ words)
-├── audios/
-│   ├── female/
-│   │   ├── speaker_1/
-│   │   ├── speaker_2/
-│   │   └── ... (11 female speakers)
-│   └── male/
-│       ├── speaker_1/
-│       ├── speaker_2/
-│       └── ... (8 male speakers)
-└── transcripts/
-    └── *.txt (7,936 transcript files)
+KenSpeech/
+├── README.md
+├── metadata.csv          # Main dataset with audio paths and transcripts
+├── transcripts_only.csv  # Additional transcripts without audio
+├── lexicon.csv           # Pronunciation dictionary (31K+ words)
+└── audio/                # Audio files
+    └── *.wav, *.mp3, *.mp4, *.m4a
 ```
 
-#### Metadata Schema (`metadata.csv`)
+## Metadata Schema
+
+The `metadata.csv` file contains:
 
 | Column | Type | Description |
 |--------|------|-------------|
-| audio_path | string | Relative path to audio file |
+| audio | string | Path to audio file (e.g., `audio/female_speaker_1_sample_261.mp4`) |
+| source_folder | string | Origin folder (`stt_dictionary` or `stt_transcripts`) |
+| gender | string | Speaker gender (`male` or `female`) |
+| speaker | string | Speaker identifier (`speaker_1`, `speaker_2`, etc.) |
 | transcript | string | Transcription text |
-| gender | string | Speaker gender (male/female) |
-| speaker_id | string | Speaker identifier (speaker_1, speaker_2, etc.) |
-| audio_format | string | Audio format (wav, mp3, mp4, m4a) |
-| sample_id | string | Sample identifier |
-
-### 3. Pronunciation Lexicon (`9sw01_swa_stt_dictionary_csv.csv`)
 
-A comprehensive **Swahili lexicon-phone dictionary** containing over **31,000 words** with their phonetic transcriptions. This lexicon uses the Swahili phoneset as defined by KenCorpus.
+### Example Record
 
-**Format:** `word,phoneme_sequence`
-
-**Example entries:**
-```
-wanapaswa,W AH N AH P AH S W AH
-wanapea,W AH N AH P EH AH
-wanapendwa,W AH N AH P EH ND W AH
-wanasema,W AH N AH S EH M AH
+```python
+{
+    'audio': 'audio/female_speaker_1_sample_261.mp4',
+    'source_folder': 'stt_dictionary',
+    'gender': 'female',
+    'speaker': 'speaker_1',
+    'transcript': 'masaa mawili kabla basi kuwasili...'
+}
 ```
 
 ---
 
 ## Usage
 
-### Loading with Metadata
+### Loading the Dataset
 
 ```python
 import pandas as pd
 from datasets import Dataset, Audio
 
-# Load the metadata
-metadata = pd.read_csv("stt_dictionary/metadata.csv")
-
-print(f"Total samples: {len(metadata)}")
-print(f"\nGender distribution:")
-print(metadata['gender'].value_counts())
+# Load metadata
+df = pd.read_csv("hf://datasets/Kencorpus/KenSpeech/metadata.csv")
 
-print(f"\nSample record:")
-print(metadata.iloc[0])
+print(f"Total samples: {len(df)}")
+print(df.head())
 ```
 
-### Loading as Hugging Face Dataset
+### Loading with Hugging Face Datasets
 
 ```python
-from datasets import Dataset, Audio
-import pandas as pd
+from datasets import load_dataset
 
-# Load metadata
-df = pd.read_csv("stt_dictionary/metadata.csv")
-
-# Create dataset
-dataset = Dataset.from_pandas(df)
-
-# Cast audio column to Audio type
-dataset = dataset.cast_column("audio_path", Audio(sampling_rate=16000))
+# Load the dataset
+dataset = load_dataset("Kencorpus/KenSpeech")
 
 # Access samples
-print(dataset[0])
+print(dataset['train'][0])
 ```
 
-### Filtering by Gender/Speaker
+### Filtering by Gender
 
 ```python
 import pandas as pd
 
-metadata = pd.read_csv("stt_dictionary/metadata.csv")
+df = pd.read_csv("metadata.csv")
 
-# Filter female speakers only
-female_samples = metadata[metadata['gender'] == 'female']
-print(f"Female samples: {len(female_samples)}")
+# Get female speakers only
+female_df = df[df['gender'] == 'female']
+print(f"Female samples: {len(female_df)}")
 
-# Filter by specific speaker
-speaker_1_samples = metadata[metadata['speaker_id'] == 'speaker_1']
-print(f"Speaker 1 samples: {len(speaker_1_samples)}")
+# Get male speakers only  
+male_df = df[df['gender'] == 'male']
+print(f"Male samples: {len(male_df)}")
 ```
 
-### Loading the Pronunciation Lexicon
+### Loading Audio with Transcripts
 
 ```python
+from datasets import Dataset, Audio
 import pandas as pd
 
-# Load the lexicon
-lexicon = pd.read_csv(
-    "KenSpeech/stt_dictionary/9sw01_swa_stt_dictionary_csv.csv",
-    header=None,
-    names=['word', 'phonemes'],
-    usecols=[0, 1]
-)
+# Load and create dataset
+df = pd.read_csv("metadata.csv")
+dataset = Dataset.from_pandas(df)
+
+# Cast audio column
+dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
 
-print(f"Lexicon contains {len(lexicon)} words")
-print(lexicon.head(10))
+# Iterate through samples
+for sample in dataset:
+    audio = sample['audio']
+    transcript = sample['transcript']
+    gender = sample['gender']
+    print(f"[{gender}] {transcript[:50]}...")
 ```
 
-### Training ASR Model (Example with Hugging Face)
+---
 
-```python
-from datasets import load_dataset, Audio
-from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
+## Pronunciation Lexicon
+
+The `lexicon.csv` file contains over 31,000 Swahili words with their phonetic transcriptions.
 
-# Load a pre-trained multilingual model
-model_name = "facebook/wav2vec2-large-xlsr-53"
-processor = Wav2Vec2Processor.from_pretrained(model_name)
-model = Wav2Vec2ForCTC.from_pretrained(model_name)
+**Format:** `word,phoneme_sequence`
 
-# Process audio files for training
-# (Additional preprocessing code would go here)
+**Example entries:**
+```
+wanapaswa,W AH N AH P AH S W AH
+wanasema,W AH N AH S EH M AH
+wanataka,W AH N AH T AH K AH
 ```
 
 ---
 
 ## Speech Types
 
-The dataset contains two types of speech:
-
-### Read Speech (96.4%)
-- Duration: 26 hours 32 minutes 37 seconds
-- Carefully articulated recordings from prepared texts
-- Higher quality and consistency
-
-### Spontaneous Speech (3.6%)
-- Duration: 59 minutes 13 seconds
-- Natural, unscripted speech
-- More representative of real-world scenarios
+| Type | Duration | Percentage |
+|------|----------|------------|
+| Read Speech | 26h 32m 37s | 96.4% |
+| Spontaneous Speech | 59m 13s | 3.6% |
 
 ---
 
 ## Intended Uses
 
-### Primary Uses
 - Training automatic speech recognition (ASR) systems for Swahili
 - Evaluating speech-to-text models
 - Phonetic and linguistic research on Swahili
 - Building text-to-speech (TTS) systems
 - Transfer learning for other Bantu languages
 
-### Out-of-Scope Uses
-- Non-speech audio processing
-- Languages other than Swahili
-- Speaker identification (speakers are anonymized)
-
----
-
-## Limitations
-
-- **Regional Focus**: Recordings are from Kenyan Swahili speakers, which may not represent all Swahili dialects (e.g., Tanzanian Swahili)
-- **Speaker Diversity**: Limited to 26 speakers with gender imbalance (19 female, 7 male)
-- **Recording Conditions**: Recordings were made in controlled environments; performance may vary in noisy conditions
-- **Spontaneous Speech**: Limited spontaneous speech data (~4% of total)
-
 ---
 
 ## Dataset Curators
@@ -251,8 +196,6 @@ The dataset contains two types of speech:
 
 ## Citation
 
-If you use this dataset in your research, please cite:
-
 ```bibtex
 @article{wanjawa2022kencorpus,
   title={Kencorpus: A Kenyan Language Corpus of Swahili, Dholuo and Luhya for Natural Language Processing Tasks},
@@ -268,8 +211,6 @@ If you use this dataset in your research, please cite:
 
 - **Research Paper**: https://arxiv.org/abs/2208.12081
 - **Dataverse**: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/KLCKL5
-- **ResearchGate**: https://www.researchgate.net/publication/371767223
-- **Semantic Scholar**: https://www.semanticscholar.org/paper/8cf70c5cd8b195ed7a399ea2cdc0b0e8f08c61ce
 
 ---
 
@@ -281,4 +222,4 @@ This dataset is licensed under **CC-BY-4.0**.
 
 ## Acknowledgments
 
-This dataset is part of the **Kencorpus** project, which aims to create NLP and speech resources for low-resource Kenyan languages. We thank all speakers and annotators who contributed to this dataset.
+This dataset is part of the **Kencorpus** project, which aims to create NLP and speech resources for low-resource Kenyan languages.