Low-Latency Voice Sales Call Bot
Budget: $10,000 – $20,000 USD
I’m building a fully-automated voice bot that can place large volumes of cold sales calls in both English and Spanish. The heart of the system will be an LLM that drives the conversation, paired with streaming speech-to-text and text-to-speech so responses feel natural and arrive with almost no delay.
Key expectations
• Real-time audio streaming: the transcription engine must keep up with live dialogue, and the TTS layer should return audio quickly enough to avoid awkward pauses.
• Clean LLM integration: the model must receive each user utterance in milliseconds, maintain short-term context, and craft persuasive sales replies that stay on script when required.
• Scalable call handling: the architecture should support bursts of concurrent outbound calls without degradation in latency or voice quality.
Deliverables
1. A deployable service (Docker or similar) that accepts a phone number list, initiates calls, and handles the full STT → LLM → TTS loop.
2. Configuration for both English and Spanish voices with seamless language switching.
3. Simple REST or gRPC hooks so I can feed new prompts or adjust the sales script on the fly.
4. Basic analytics—call duration, hang-up reason, and transcript export.
5. Instructions for running the stack on a cloud instance of my choice.
Acceptance criteria
• End-to-end response time under two seconds for 95 % of turns.
• Minimum 90 % transcription accuracy on common phone-line audio.
• Demonstrated parallel handling of at least 50 simultaneous calls in a controlled test.
If you have experience combining SIP or Twilio dialing with Whisper-style STT, real-time TTS engines like Amazon Polly or ElevenLabs, and conversational LLMs, I’d love to see what you can bring to the project.
Deadline: Within 10 days.
Key expectations
• Real-time audio streaming: the transcription engine must keep up with live dialogue, and the TTS layer should return audio quickly enough to avoid awkward pauses.
• Clean LLM integration: the model must receive each user utterance in milliseconds, maintain short-term context, and craft persuasive sales replies that stay on script when required.
• Scalable call handling: the architecture should support bursts of concurrent outbound calls without degradation in latency or voice quality.
Deliverables
1. A deployable service (Docker or similar) that accepts a phone number list, initiates calls, and handles the full STT → LLM → TTS loop.
2. Configuration for both English and Spanish voices with seamless language switching.
3. Simple REST or gRPC hooks so I can feed new prompts or adjust the sales script on the fly.
4. Basic analytics—call duration, hang-up reason, and transcript export.
5. Instructions for running the stack on a cloud instance of my choice.
Acceptance criteria
• End-to-end response time under two seconds for 95 % of turns.
• Minimum 90 % transcription accuracy on common phone-line audio.
• Demonstrated parallel handling of at least 50 simultaneous calls in a controlled test.
If you have experience combining SIP or Twilio dialing with Whisper-style STT, real-time TTS engines like Amazon Polly or ElevenLabs, and conversational LLMs, I’d love to see what you can bring to the project.
Deadline: Within 10 days.