Merge branch 'main' of https://huggingface.co/datasets/danielrosehill/Voice-Note-Audio

README.md

@@ -0,0 +1,132 @@
+---
+task_categories:
+- automatic-speech-recognition
+language:
+- en
+pretty_name: "Voice Note Audio"
+size_categories:
+- "n<1K"
+tags:
+- speech-to-text
+- noise-robustness
+- evaluation
+- whisper
+license: mit
+---
+
+# Voice Notes
+
+A dataset of voice notes collected by Daniel Rosehill in and around Jerusalem (mostly) in a variety of acoustic environments and in a variety of formats reflecting typical daily use with speech to text transcription apps.
+
+This dataset is a subsection of a voice note training dataset that I'm curating for STT fine-tuning and entity recognition.
+
+## Annotation
+
+The dataset includes rich annotations collected using Label Studio:
+
+- Corrected transcripts (manually corrected AI transcripts)
+- Audio quality ratings
+- Environmental information (recording location, microphone type, etc.)
+- Content classification
+- Audio challenges present
+- Language information
+- Entity recognition
+- Audio source identification
+
+## Label Studio Configuration Parameters
+
+### Audio Challenges Present
+Multiple selection options for identifying audio quality issues:
+- Traffic Noise: Road traffic sounds
+- Audible Conversations: Other people talking
+- Outdoor Noise (General): Street/urban sounds
+- Background Music: Music playing
+- **Crying Baby**: Baby crying sounds (newly added)
+
+### Incidental Audio Pickup Source
+Single selection for identifying the source of incidental audio:
+- **Speaker**: Audio from the primary speaker
+- **Others**: Audio from other sources
+
+### Background Conversation Language
+Single selection for identifying the language of background conversations:
+- **English**
+- **Hebrew** 
+- **Arabic**
+- **French**
+- **Russian**
+
+### Multilingual Transcript
+Single selection to indicate if the transcript contains multiple languages:
+- **True**: Transcript contains multiple languages
+- **False**: Transcript is in a single language
+
+### Entities Present in Note
+Multiple selection for identifying named entities mentioned in the voice note:
+- **Dates**: Specific dates or time references
+- **Persons**: Names of people
+- **Placenames**: Geographic locations or places
+- **Email Addresses**: Email addresses mentioned
+- **Blog Title**: Blog or article titles
+- **Acronym**: Acronyms or abbreviations
+- **Organisations**: Company or organization names
+
+### Bluetooth Codec
+Single selection for identifying the Bluetooth codec used during recording:
+- **SBC**: Standard Bluetooth codec
+- **AAC**: Advanced Audio Coding
+- **aptX**: Qualcomm aptX codec
+- **aptX HD**: High-definition aptX codec
+- **LDAC**: Sony LDAC high-quality codec
+- **LC3**: Low Complexity Communication Codec
+- **N/A**: Not applicable (wired/internal mic)
+- **Unknown**: Codec information unavailable
+
+## Microphones Used
+
+The voice notes in this dataset were recorded using various microphones:
+- **OnePlus Nord 3 Internal Microphone**: Built-in phone microphone
+- **Poly 5200**: Bluetooth-connected microphone
+- **ATR 4697**: Professional microphone
+
+## Data Organization
+
+- `audio/` - Processed audio files (MP3/WAV)
+- `transcripts/` - Transcript files
+  - `uncorrected/` - AI-generated transcripts
+  - `ground_truths/` - Manually corrected transcripts (ground truth)
+- `annotations/` - Annotation task files and completed annotations
+- `candidate-parameters.md` - Additional parameters for future implementation
+- `preprocessing/` - Workflow for adding new data (see preprocessing/README.md) 
+
+## Purpose
+
+This collection, consisting of voice notes recorded by Daniel Rosehill using Voicenotes.com, is specifically gathered to evaluate and improve the robustness of Speech-to-Text (STT) systems under non-ideal, real-world conditions. Unlike studio-quality audio used for training, these notes often contain various types of background noise, overlapping conversations, and environmental distortions typical of everyday recording scenarios.
+
+This dataset serves three primary objectives:
+
+### 1. Personal STT Fine-Tuning
+Improve speech recognition accuracy for personal voice notes by creating a refined transcription model tailored to individual speech patterns and common recording environments.
+
+### 2. Voice Note Entity Recognition
+Develop a specialized model for the "Voice Router" application to classify and identify entities within voice note recordings, enabling intelligent routing and categorization of voice-based content.
+
+### 3. Public Research Dataset
+Generate a comprehensive, open-source dataset with rich annotations for various audio recording conditions, enabling STT model evaluation across different acoustic environments and contributing to the broader speech recognition research community.
+
+The dataset contains approximately 700 voice notes totaling 13 hours of audio. Each audio file comes with an AI-generated transcript provided by Voicenotes.com's STT service, serving as a baseline for comparison. A subset of these transcripts will be manually corrected to create a high-quality ground truth dataset for fine-tuning STT models and developing a comprehensive, nuanced speech recognition research and development framework focused on real-world voice note transcription challenges.
+
+## Contents
+
+- `audio/`: Folder containing the original MP3 audio files of the voice notes.
+- `transcripts/`: Folder containing transcript files
+  - `uncorrected/`: Raw, AI-generated transcripts corresponding to the audio files
+  - `ground_truths/`: Manually corrected transcripts for training and evaluation
+- `dataset_metadata.json`: Metadata associated with the dataset entries.
+- `label_studio_config.xml`: Configuration file for Label Studio, an annotation tool.
+- `setup_annotation.py`: Script to help set up the annotation process.
+- `parameters.md`: A detailed list of parameters to be annotated for each voice note.
+
+## Annotation
+
+The `parameters.md` file specifies the key aspects to be annotated for each voice note, including audio quality, speaker characteristics, transcription accuracy, and contextual information. This structured annotation will provide valuable metadata for analyzing STT performance and guiding model improvements.