What is the best speech-to-text API for sales call recording and analysis?

AssemblyAI's Universal-3 Pro is the leading choice for sales call recording and analysis — 94%+ transcription accuracy on noisy phone audio, 43% better alphanumeric accuracy than competing providers, and built-in entity detection that captures product names, pricing tiers, and competitor mentions. For real-time conversation intelligence during live calls, Universal-3 Pro Streaming delivers ~150ms P50 latency at $0.45/hr per stream with unlimited concurrency. Pair it with the LLM Gateway (25+ models across Claude, GPT, and Gemini through one unified API) to produce coaching scorecards, sentiment shift analysis, and deal risk scoring. For a fully managed conversational agent — lead qualification, appointment booking, outbound dialing — the Voice Agent API bundles STT, LLM, and TTS at $4.50/hr per session.

How do I build a sales voice agent that qualifies leads and books meetings?

Start with AssemblyAI's Voice Agent API for the fastest path to a working agent (~1s end-to-end). Configure a system prompt that defines your qualification criteria (BANT: budget, authority, need, timeline) and the agent's tools — calendar lookup, CRM enrichment, meeting creation. The Voice Agent API handles STT, LLM, and TTS over one WebSocket at $4.50/hr per session. Add product names, pricing tiers, and SKUs as keyterms so the model recognizes them in conversation. For BYO orchestration, use Universal-3 Pro Streaming ($0.45/hr) inside LiveKit, Pipecat, or Vapi with your own LLM driving the qualification logic. Either path connects to telephony via Twilio Media Streams or SIP — most teams ship a working agent in an afternoon.

How does conversation intelligence work for sales teams?

Conversation intelligence captures every sales call as structured data: speaker-separated transcript, sentiment timeline, entity detection (competitor mentions, pricing objections, product names), and LLM-generated coaching scorecards. AssemblyAI's Universal-3 Pro Streaming ($0.45/hr) provides the real-time STT and analytics layer; the LLM Gateway (25+ models across Claude, GPT, and Gemini) produces talk/listen ratios, sentiment shift analysis, and specific coaching suggestions per rep. The pipeline pushes summaries, action items, and deal risk scores into Salesforce, HubSpot, or any CRM via webhook — so sales managers don't have to review every call to know which ones need attention.

What metrics can I extract from sales calls with AI?

AssemblyAI's analytics layer produces several metrics out of the box: talk/listen ratio (how much the rep speaks vs the customer), sentiment timeline (emotional shifts throughout the call), entity detection (competitor mentions, pricing objections, product names, decision-maker references), topic detection (which discovery questions got asked), and call outcome classification (booked / no-decision / lost). Use the LLM Gateway to layer on derived metrics — BANT scoring, deal risk, follow-up commitments, coaching suggestions per rep — using your own prompt templates. All metrics are available on the live stream during the call (for real-time agent assist) and as a post-call summary pushed to your CRM.

Can I use AI voice agents for outbound sales calls?

Yes. AssemblyAI's Voice Agent API handles full outbound voice automation — dial out via Twilio or your telephony provider, run a sales agent with system-prompted qualification logic, and either book a meeting or transfer to a live rep when the lead qualifies. One WebSocket replaces separate STT, LLM, and TTS providers at $4.50/hr per session, with sub-1-second end-to-end latency. The agent captures conversation context (BANT signals, objections raised, next steps committed) and pushes structured data to your CRM after the call. For teams that already run LiveKit, Pipecat, or Vapi for orchestration, Universal-3 Pro Streaming ($0.45/hr) drops in as the STT layer with your own LLM driving the outbound script.

How do I integrate sales call intelligence with my CRM?

AssemblyAI's pipeline pushes structured call data to your CRM via webhook — Salesforce, HubSpot, Pipedrive, Close, or any custom system. After each call you receive a summary, action items, sentiment scores, deal risk, and competitive intelligence (which competitors got mentioned and in what context). Use PII redaction (redact_pii=true) to mask names, card numbers, and account IDs before transcripts hit the CRM if your compliance posture requires it. Voice Agent API sessions can also fire webhooks during the call (e.g., mid-call notes from tool calls) so reps see context updates in real time without waiting for the post-call summary.

Solutions

Voice agents for sales & call intelligence

Build AI voice agents that qualify leads, book meetings, and handle outbound calls — then analyze every conversation with sentiment analysis, speaker diarization, and LLM-powered coaching scorecards. Powered by the fastest, most accurate speech-to-text.

Get started free Contact sales

Sales call scorecard

Rep: Sarah K. · Prospect: Acme Corp · 18 min

Talk / Listen

38 / 62

Sentiment

Positive

Sentiment timeline

0:00 18:42

Coaching suggestions

1. Strong discovery — keep asking about timeline
2. Address pricing objection earlier
3. Confirm next steps before ending call

The problem

Your reps are flying blind after every call

Sales managers review less than 2% of calls. Reps self-report outcomes. Competitor mentions, pricing objections, and buying signals vanish the moment the call ends. Meanwhile, manual lead qualification wastes hours that should be spent selling. Modern voice agents and AI-powered call intelligence — built on accurate streaming STT, a managed LLM, and sentiment analysis — capture everything and turn every conversation into structured, coachable data.

Built for revenue team performance

Latency ~150ms

P50 median streaming latency for Universal-3 Pro Streaming.

Entity 43%

Better alphanumeric accuracy than other providers.

Uptime 99.9%

SLA with SOC 2 Type 2 certification.

Scale 40TB+

Audio processed daily in production.

Two ways to build

Pick the API that fits your sales stack

Ship an AI sales agent in an afternoon, or drop industry-leading STT into the conversation intelligence platform you already run.

Recommended

Voice Agent API

Our proprietary voice stack via one WebSocket. Build lead qualification agents, appointment setters, and outbound dialers with zero infra to manage.

Best for

Lead qualification, appointment booking, outbound calling
Teams shipping fast — working agent in an afternoon
Keyterm boosting for product names and pricing tiers
Claude Code compatible — paste the docs and build anything

$4.50/hr — speech, LLM, and voice all included

Get started for free

Free tier available · No credit card required

Bring Your Own Stack

Universal-3 Pro Streaming STT API

The STT and analytics layer for your conversation intelligence platform. Works natively with your preferred orchestrator and CRM integration.

Best for

Teams using LiveKit, Pipecat, or Vapi as their orchestrator
Conversation intelligence with sentiment + entity detection
LLM Gateway for coaching scorecards and deal risk scoring
PII redaction before transcripts hit your CRM or warehouse
High-scale deployments where margin and full control matter

$0.45/hr — transcription only, unlimited streams

View integration docs

No concurrency caps · Autoscaling included

Your sales intelligence pipeline

Ingest call audio

Voice Agent API: single WebSocket for live agents. Or connect recordings from your dialer, Twilio, or call center platform.

Transcribe with speaker diarization

Speaker labels separate rep and customer. Sentiment analysis tracks emotional shifts. Entity detection catches competitor mentions and pricing.

Generate coaching scorecards

LLM Gateway produces talk/listen ratios, sentiment shift analysis, and specific coaching suggestions per rep. 25+ models across Claude, GPT, and Gemini.

Push to CRM

Summaries, action items, and deal risk scores pushed to Salesforce, HubSpot, or any CRM via webhook.

analytics

Call intelligence

Talk ratio

38%

Listen ratio

62%

Sentiment timeline

0:00 18:42

Competitor: Gong Buying signal

Quickstart

Get a working agent in minutes

Voice Agent API — recommended

# Voice Agent API: sales lead qualification agent
import asyncio, json, websockets

API_KEY = "YOUR_API_KEY"

async def run_agent():
    async with websockets.connect(
        "wss://agents.assemblyai.com/v1/ws",
        additional_headers={"Authorization": f"Bearer {API_KEY}"},
    ) as ws:
        await ws.send(json.dumps({
            "type": "session.update",
            "session": {
                "system_prompt": (
                    "You are a sales qualification agent for Acme Corp. "
                    "Ask about budget, timeline, and decision-maker."
                ),
                "greeting": "Hi, thanks for your interest in Acme — how can I help?",
                "input": {"keyterms": ["Acme Pro", "Enterprise Plan", "tier-2"]},
                "output": {"voice": "ivy"},
            },
        }))
        # Stream audio in, get audio + transcript back
        async for msg in ws:
            handle(json.loads(msg))  # transcript.user, reply.audio, tool.call, ...

Universal-3 Pro Streaming + LiveKit — BYO stack

# LiveKit + AssemblyAI STT in a cascading sales agent pipeline
from livekit.agents import Agent, AgentSession, TurnHandlingOptions
from livekit.plugins import assemblyai, cartesia, openai, silero

class SalesAgent(Agent):
    def __init__(self):
        super().__init__(
            instructions=(
                "You are a sales qualification agent for Acme Corp. "
                "Be concise. Qualify on budget, timeline, and authority."
            ),
        )

async def entrypoint(ctx):
    session = AgentSession(
        stt=assemblyai.STT(
            model="u3-rt-pro",
            min_turn_silence=100,
            max_turn_silence=1000,                # let buyers finish thoughts on email/phone numbers
            vad_threshold=0.3,
            keyterms_prompt=["Acme Pro", "Enterprise Plan", "tier-2"],
        ),
        llm=openai.LLM(model="gpt-4o"),
        tts=cartesia.TTS(),
        vad=silero.VAD.load(activation_threshold=0.3),
        turn_handling=TurnHandlingOptions(
            turn_detection="stt",
            endpointing={"min_delay": 0},          # avoid additive latency in STT-driven turns
        ),
    )
    await session.start(room=ctx.room, agent=SalesAgent())

Try in Playground View full docs

Resolution-grade accuracy

Universal-3 Pro Streaming transcribes 94%+ on noisy phone audio — the difference between a captured competitor mention and a missed coaching opportunity.

PII redaction by default

Names, card numbers, addresses, and account IDs masked before transcripts hit your CRM, data warehouse, or QA stack.

Real-time intelligence

Topic detection, sentiment, and call outcomes available on the live stream — coach agents in the moment, not the next day.

Teams shipping sales intelligence on AssemblyAI

25% reduction in handling time

EdgeTier's customer CarTrawler reduced chat handling time by 25% through enhanced insights and agent optimization powered by AssemblyAI.

EdgeTier

The accuracy was strong, but the great documentation and unique models like Auto Chapters and Sentiment Analysis is what really won us over.

Nathan Webb, Product Manager — Aloware

: AssemblyAI's Universal-3 Pro is the leading choice for sales call recording and analysis — 94%+ transcription accuracy on noisy phone audio, 43% better alphanumeric accuracy than competing providers, and built-in entity detection that captures product names, pricing tiers, and competitor mentions. For real-time conversation intelligence during live calls, Universal-3 Pro Streaming delivers ~150ms P50 latency at $0.45/hr per stream with unlimited concurrency. Pair it with the LLM Gateway (25+ models across Claude, GPT, and Gemini through one unified API) to produce coaching scorecards, sentiment shift analysis, and deal risk scoring. For a fully managed conversational agent — lead qualification, appointment booking, outbound dialing — the Voice Agent API bundles STT, LLM, and TTS at $4.50/hr per session.
: Start with AssemblyAI's Voice Agent API for the fastest path to a working agent (~1s end-to-end). Configure a system prompt that defines your qualification criteria (BANT: budget, authority, need, timeline) and the agent's tools — calendar lookup, CRM enrichment, meeting creation. The Voice Agent API handles STT, LLM, and TTS over one WebSocket at $4.50/hr per session. Add product names, pricing tiers, and SKUs as keyterms so the model recognizes them in conversation. For BYO orchestration, use Universal-3 Pro Streaming ($0.45/hr) inside LiveKit, Pipecat, or Vapi with your own LLM driving the qualification logic. Either path connects to telephony via Twilio Media Streams or SIP — most teams ship a working agent in an afternoon.
: Conversation intelligence captures every sales call as structured data: speaker-separated transcript, sentiment timeline, entity detection (competitor mentions, pricing objections, product names), and LLM-generated coaching scorecards. AssemblyAI's Universal-3 Pro Streaming ($0.45/hr) provides the real-time STT and analytics layer; the LLM Gateway (25+ models across Claude, GPT, and Gemini) produces talk/listen ratios, sentiment shift analysis, and specific coaching suggestions per rep. The pipeline pushes summaries, action items, and deal risk scores into Salesforce, HubSpot, or any CRM via webhook — so sales managers don't have to review every call to know which ones need attention.
: AssemblyAI's analytics layer produces several metrics out of the box: talk/listen ratio (how much the rep speaks vs the customer), sentiment timeline (emotional shifts throughout the call), entity detection (competitor mentions, pricing objections, product names, decision-maker references), topic detection (which discovery questions got asked), and call outcome classification (booked / no-decision / lost). Use the LLM Gateway to layer on derived metrics — BANT scoring, deal risk, follow-up commitments, coaching suggestions per rep — using your own prompt templates. All metrics are available on the live stream during the call (for real-time agent assist) and as a post-call summary pushed to your CRM.
: Yes. AssemblyAI's Voice Agent API handles full outbound voice automation — dial out via Twilio or your telephony provider, run a sales agent with system-prompted qualification logic, and either book a meeting or transfer to a live rep when the lead qualifies. One WebSocket replaces separate STT, LLM, and TTS providers at $4.50/hr per session, with sub-1-second end-to-end latency. The agent captures conversation context (BANT signals, objections raised, next steps committed) and pushes structured data to your CRM after the call. For teams that already run LiveKit, Pipecat, or Vapi for orchestration, Universal-3 Pro Streaming ($0.45/hr) drops in as the STT layer with your own LLM driving the outbound script.
: AssemblyAI's pipeline pushes structured call data to your CRM via webhook — Salesforce, HubSpot, Pipedrive, Close, or any custom system. After each call you receive a summary, action items, sentiment scores, deal risk, and competitive intelligence (which competitors got mentioned and in what context). Use PII redaction (redact_pii=true) to mask names, card numbers, and account IDs before transcripts hit the CRM if your compliance posture requires it. Voice Agent API sessions can also fire webhooks during the call (e.g., mid-call notes from tool calls) so reps see context updates in real time without waiting for the post-call summary.

Build your sales voice agent today

Free tier, no credit card. From lead qualification to coaching scorecards in an afternoon.

Get started free