Skip to main content
ASR-optimized noise cancellation for voice AI pipelines
Model Code: AGENTIC_NC
Sample Rate: 16 kHz
Use Case: Voice agents, IVR systems, ASR pipelines

Introduction

AI Voice Agent is a specialized noise cancellation model designed to improve Automatic Speech Recognition (ASR) accuracy in noisy environments. Unlike traditional noise cancellation optimized for human perception, this model preserves the acoustic and phonetic information that ASR systems depend on while removing conversational disruptions.

Key Advantages

FeatureAI Voice AgentTraditional Noise Cancellation
Optimization GoalASR accuracy (WER reduction)Human listening quality
Acoustic PreservationPreserves phonetic featuresMay strip ASR-critical info
ASR CompatibilityAll major ASR systemsMay degrade ASR performance
WER Improvement5-30% in noisy conditionsOften increases WER
Clean Audio ImpactNo degradationNo degradation

Key Benefits

Improved ASR Accuracy

5-30% average WER improvement
across multiple ASR systems on noisy data with no degradation on clean audio.

ASR-Agnostic Design

Works seamlessly with any ASR pipeline—open-source or commercial—without requiring retraining or modification.

Enhanced Turn-Taking

Reduces false triggers from ambient sounds (background chatter, environmental noise) that cause agents to interrupt.

Real-World Robustness

Trained on diverse acoustic environments to handle production variability from call centers to mobile devices.

Use Cases

Use CaseBenefitsIdeal For
Voice AgentsImproves speech recognition in noisy environments; preserves the primary speaker onlyConversational AI, virtual assistants, chatbots
IVR SystemsEnhances recognition accuracy for automated phone systemsInteractive voice response, phone menus, call routing
Customer Service BotsReduces transcription errors in call center environmentsSupport automation, call analysis, and quality monitoring
Phone AssistantsOptimizes speech-to-text in telephony conditionsMobile assistants, hands-free applications
Real-Time TranscriptionClean audio input for live transcription servicesMeeting transcription, live captioning, and note-taking
ASR PreprocessingPurpose-built enhancement for downstream ASR systemsAny speech-to-text pipeline in noisy conditions

Model Configuration

To use the AI Voice Agent model, specify it when creating your audio processor:
audio_params.modelName = "AGENTIC_NC"
audio_params.sampleRate = 16000  # Up to 16kHz supported
  • Sample Rate: 16 kHz for best quality
Note: Audio with sample rates > 16 kHz will be downsampled.
For complete setup instructions, see the Quickstart Guide.

Performance Benchmarks

Real-World Test Results

Test Environment:

Background: Restaurant with cutlery sounds and conversation Microphone: Speakerphone at 10cm distance ASR System: Deepgram Nova3 Streaming Transcription Accuracy & Audio Samples
Oracle TranscriptSource TranscriptGeneric BVCSanas AI Voice Agent
it was in the spring of the year eighteen ninety four that all london was interested and the fashionable world dismayed by the murder of the honorable ronald adair under most unusual and inexplicable circumstances the public has already learned those particulars of the crime which came out in the police investigation but a good deal was suppressed upon that occasion since the case for the prosecution was so overwhelmingly strong that it was not necessary to bring forward all the facts only now at the end of nearly ten years am i allowed to supply those missing links which make up the whole of the remarkable chain the crime was of interest in itself but that interest was as nothing to me compared to the incons inconceivable sequence which afforded me the greatest shock and surprise of any event in my adventurous lifeit was in the spring of the year one eight nine four that all london was interested and the fashionable world dismayed by the mother of the honorable rona adair under most unusual and those particulars of the crime which came out in the police investigation but a good deal was suppressed upon the occasion since the case for the prosecution strong that it was not necessary necessary to bring forward all the facts only now at the end of nearly ten years am i allowed to supplement these basic things which make the crime was of interest in itself but that interest was as nothing to me compared to the incon inconceivable sequence which afforded me the greatest shock and of any event in my adventurous life let’s create magicit was in the one eight nine four that all london was interested and the fashionable world dismayed by the mother of the honorable rona adair and the most unusual and inexplicable circumstances the public has already learned those particulars of the crime which came out in the police investigation but a good deal was suppressed upon the occasion since the case for the prosecution was so overwhelmingly strong that it was not necessary to bring forward all the facts only now at the end of nearly ten years number allowed to supply these basic wings which make up the board of the remarkable chain the crime was of interest in itself but that interest was as nothing to me compared to the incomes inconsistencyable sequence which afforded me the greatest shock and surprise of any event in my adventurous lifeit was in the spring of the year one eight nine four that all london was interested and the fashionable world dismayed by the murder of the honorable rona adair and the most unusual and inexplicable circumstances the public has already learned those particulars of the crime which came out in the police investigation but a good deal was surprised upon the occasion since the case for the prosecution was so overwhelmingly strong that it was not necessary to bring forward all the facts only now at the end of nearly ten years am i allowed to supply these missing links of which make up the whole of the remarkable chain the crime was of interest in itself but that interest was as nothing to me compared to the inconceivable sequence which afforded me the greatest shock and surprise of any event in my adventurous life
22.9% WER13.2% WER7.5% WER

Latency & Performance

MetricValueDescription
Streaming Latency~100msEnd-to-end processing time.

Note: Actual latency also depends on the selected Sanas Cloud region and network conditions.
ConcurrencyScalableLimited only by your infrastructure.

Next Steps

Need help? Contact our support team at support@sanas.ai or raise a ticket.