Deepgram
Your AI agent gains ears and a voice. Deepgram transcribes customer call recordings in real time, summarizes audio content, and converts text to natural-sounding speech. Build voice-enabled support that handles inquiries across 30+ languages without a human on the line.
From transcribing customer calls to generating audio responses, your AI agent taps into Deepgram's Nova models to process voice data during live interactions.
Deepgram
Real examples of how businesses use Deepgram-powered AI agents to convert spoken customer interactions into actionable data and automated responses.
A customer leaves a voicemail at midnight about a billing dispute. Your AI Agent picks up the recording, sends it through Deepgram's Speech-to-Text API with smart formatting and diarization, and produces a clean transcript. The agent then categorizes the issue, creates a support ticket with the full transcript attached, and prioritizes it for the morning shift. Your team starts the day with organized, actionable voicemail data instead of a cluttered inbox.
A support center handles calls in Spanish, French, and English. After each call, the AI Agent sends the recording to Deepgram's Summarize Audio endpoint with automatic language detection. The agent extracts a concise summary of the customer's issue and resolution, then logs it in your CRM. Managers across regions review standardized summaries without listening to full recordings, saving hours of manual review every week.
A visually impaired customer interacts with your website chat. Your AI Agent composes the answer and converts it to audio using Deepgram's Text-to-Speech API with the Aura voice model. The customer hears a natural spoken response instead of reading text. Your business delivers inclusive, accessible support that serves every customer, regardless of how they prefer to consume information.
Deepgram
FAQs
The agent defaults to Deepgram's Nova-3 model, which delivers a 53% lower word error rate than competing services. You can also configure it to use specialized models like the phonecall model for call center audio or the general model for broader use cases. The model parameter is adjustable per request.
Yes. The agent enables Deepgram's diarize parameter when processing multi-speaker recordings. Each word in the transcript gets assigned a speaker label, so your team can see exactly who said what. This works with all Deepgram ASR models including Nova and is included at no extra charge.
Deepgram supports WAV, MP3, FLAC, OGG, and most common audio formats. The content_type parameter specifies the MIME type. Language support covers 30+ languages including English, Spanish, French, German, Portuguese, and Mandarin. Enable detect_language for automatic language identification.
The agent calls Deepgram's TTS REST API with the text content, selecting from available Aura voice models. You can customize pitch (0.5 to 2.0), speed (0.25 to 4.0), output format (MP3, WAV), and sample rate. The audio file is generated and delivered to the customer within the conversation thread.
Tars does not store your audio files. Audio URLs are sent to Deepgram for processing, and the resulting transcripts or summaries are used within the active conversation. Deepgram processes audio in-memory by default. You control data retention policies through your Deepgram project settings.
Topic Detection transcribes the audio and then identifies the main subjects discussed using Deepgram's audio intelligence layer. The agent receives a list of detected topics with confidence scores. This enables automatic categorization of customer calls, such as tagging billing inquiries, technical issues, or sales conversations without manual review.
Deepgram's Nova-3 model is specifically optimized for noisy, real-world audio. You can boost accuracy further using the keywords parameter to prioritize industry-specific terms like product names or technical jargon. The enhanced tier option provides additional accuracy for challenging audio quality.
Yes. The agent can query Deepgram's Usage Summary API to retrieve minutes submitted, processed, and billed for any time range. Filter by model, accessor, or tag to break down usage across different agent workflows. This helps you monitor costs and optimize which audio gets processed.
Don't limit your AI Agent to basic conversations. Watch how to configure and add powerful tools making your agent smarter and more functional.
Privacy & Security
At Tars, we take privacy and security very seriously. We are compliant with GDPR, ISO, SOC 2, and HIPAA.
Email - sales@hellotars.com
.png)
%201.png)
%201.png)
