--- license: cc-by-4.0 --- ## Dataset Description This dataset is a large-scale collection of **127,624 hours of processed English (US) single-channel call center audio recordings**, derived from a broader corpus of **2,065,026 hours of processed call center audio data spanning 32 languages**, designed for the development, training, and evaluation of advanced Speech AI, Conversational AI, ASR, STT, SFT, and RLHF systems. The dataset consists of real-world customer service and support conversations collected from operational call center environments. Audio is provided in a **single-channel format**, where conversational speech is captured as a unified audio stream, closely reflecting real-world telephony deployments and production speech processing pipelines. This dataset is particularly valuable for building scalable enterprise-grade AI systems including Automatic Speech Recognition (ASR), Speech-to-Text (STT), Call Analytics, Conversational Intelligence, Sentiment Analysis, Large Language Model (LLM) training pipelines, Supervised Fine-Tuning (SFT), and Reinforcement Learning with Human Feedback (RLHF). --- ## Audio Processing & Refinement Pipeline To ensure enterprise-grade quality and usability, the dataset undergoes a comprehensive 4-step audio refining and processing pipeline before final delivery: ### 1. Duplicate Asset Elimination Removal of duplicate or repeated recordings to maintain dataset uniqueness, consistency, and high-quality training data. ### 2. Low-Activity Voice Removal Filtering of silent, low-volume, inactive, or low-quality audio samples to improve overall dataset reliability. ### 3. PII Detection & Muting Automatic detection and redaction/muting of personally identifiable information (PII) to support privacy compliance and safe AI training. ### 4. Background Noise Removal Application of advanced noise-reduction and audio-cleaning techniques to enhance speech clarity and improve model performance. This processing pipeline ensures that the dataset is clean, scalable, production-ready, and optimized for speech AI, conversational AI, ASR, SFT, and RLHF workflows. --- # Dataset Specification * Duration: 127,624 Hours * Language: English (US) * Type: Processed * Audio Conditions: Real-world call center environments * Channel Configuration: Single Channel * Format: .wav, .mp3, .ogg, etc. * Sampling Rate: 8000 Hz --- # Key Use Cases * Automatic Speech Recognition (ASR) * Speech-to-Text (STT) Systems * Call Center AI Applications * Contact Center Analytics * Conversational AI Training * Large Language Model (LLM) Fine-Tuning * Supervised Fine-Tuning (SFT) * Reinforcement Learning with Human Feedback (RLHF) * Sentiment Analysis * Emotion Detection * Quality Monitoring * Compliance Monitoring * Customer Experience Analytics * Virtual Customer Support Agents * Voice AI Applications --- # Value of Single Channel Dataset * Closely reflects real-world telephony and call center deployments * Ideal for speech recognition and transcription model training * Supports conversational AI and dialogue understanding systems * Captures natural customer service interaction patterns * Suitable for large-scale AI training pipelines * Effective for sentiment and behavioral analytics * Useful for enterprise contact center intelligence platforms * Supports downstream NLP and LLM workflows --- # Audio Quality Analysis ## Signal Quality Analysis (Signal QA) To ensure robust signal-level integrity and consistency, the dataset was evaluated using multiple acoustic and signal-processing metrics. | Metric | Value | Interpretation | | --------------------- | ------ | ------------------------------------------------------------------------------- | | Average SNR (dB) | 50.03 | High signal-to-noise ratio indicating clean audio with minimal background noise | | Average RMS Energy | 0.089 | Stable signal energy level suitable for speech processing | | Silence Ratio | 0.448 | Reflects natural conversational pauses | | Clipping Ratio | 0.0 | No clipping detected, ensuring distortion-free audio | | Loudness (LUFS) | -22.12 | Well-balanced loudness within acceptable speech ranges | | Overall Quality Score | 70.83 | Good signal quality suitable for training and evaluation | --- ## DNSMOS Evaluation To ensure production-level reliability, the dataset was evaluated using DNSMOS (Deep Noise Suppression Mean Opinion Score). | Metric | Score | Interpretation | | ---------------------- | ----- | --------------------------------------------------------- | | Speech Quality (SIG) | 3.89 | Clear and intelligible conversational speech | | Background Noise (BAK) | 4.01 | Strong noise suppression and acoustic clarity | | Overall MOS (OVR) | 3.81 | High-quality real-world audio suitable for model training | --- ## SQUIM-Based Audio Quality Analysis To further assess perceptual and signal characteristics, the dataset was evaluated using SQUIM-based metrics. | Metric | Value | Interpretation | | ------------------------ | ----- | ---------------------------------------------------------- | | Average Energy | 0.003 | Controlled signal amplitude without distortion | | Spectral Flatness | 0.052 | Speech-dominant signal with minimal noise characteristics | | Zero Crossing Rate (ZCR) | 0.062 | Consistent with voiced speech and low high-frequency noise | | Dynamic Range | 1.683 | Natural conversational amplitude variation | | SI-SDR Proxy | 15.0 | Good signal-to-distortion ratio | | SQUIM Score | 62.59 | Strong perceptual quality for real-world applications | --- ## Key Insight The dataset maintains strong acoustic quality despite real-world call center conditions, making it suitable for production-grade Speech AI systems, Conversational AI platforms, LLM pipelines, speech understanding models, and enterprise customer support applications. --- # Dataset Validation via End-to-End Model Training To validate dataset effectiveness, a complete speech-to-NLP training pipeline was built and executed using InfoBay.AI audio data. ## Full Pipeline OpenAI Whisper Transcription → Sentiment Labeling → DistilBERT Training → 3-Class Sentiment Classification ### Validation Insight This end-to-end workflow demonstrates that the dataset is not only large-scale but also self-sufficient for developing downstream AI applications without reliance on external datasets. --- # Sentiment Classification Task The dataset supports supervised learning for sentiment understanding across three classes: * Negative (Class 0) * Neutral (Class 1) * Positive (Class 2) The dataset contains naturally occurring emotional and contextual variation, making it highly suitable for: * RLHF Preference Modeling * Emotion-Aware Conversational Agents * Human-Aligned Response Generation * Customer Experience Analytics * Voice-Based Sentiment Intelligence --- # Model Performance (From-Scratch Training) A DistilBERT-based model trained from scratch achieved strong performance using transcripts generated from this dataset. * Accuracy: ~98% * Macro F1 Score: ~0.98 * Weighted F1 Score: ~0.99 | Class | Sentiment | Precision | Recall | F1-score | Support | | ----- | --------- | --------- | ------ | -------- | ------- | | 0 | Negative | 0.97 | 0.96 | 0.96 | 1,128 | | 1 | Neutral | 0.99 | 0.99 | 0.99 | 7,865 | | 2 | Positive | 0.98 | 0.98 | 0.98 | 2,658 | --- # Basic JSON Schema ```json { "id": "string", "audio_filepath": "string", "duration": "float", "language": "string", "sample_rate": "integer", "format": "string", "num_speakers": "integer", "domain": "call_center", "metadata": { "source": "string", "recording_condition": "string" } } ``` **Data Creation** Procured through formal agreements and generated in the ordinary course of business. **Considerations** This dataset is provided for research and educational purposes only. It contains only sample data. For access to the full dataset and enterprise licensing options, please visit our website [InfoBay.AI](https://infobay.ai/) or contact us directly. -Ph: (91) 8303174762 -Email: datareq@infobay.ai