Kabir Goel
Articles by Kabir Goel
What is jitter in VoIP? How to measure and fix it
Jitter is the variation in when voice packets arrive. What causes it, what it sounds like, the 30 ms rule of thumb, and how to fix choppy calls, on a phone system or a voice agent.
Retell AI pricing 2026: per-minute costs and plans
What Retell AI costs in 2026: the $0.07 to $0.31 per-minute range, the voice, LLM, and telephony components behind it, add-on fees, and worked examples.
What is CPaaS? The communications API layer
CPaaS is the API model for putting calls, texts, and verification inside your own app. What it covers, how it's priced, and where voice agents fit.
Call center metrics: FCR, AHT, and what matters
First call resolution, average handle time, transfer rate, and CSAT: the formulas, current benchmarks, and how the math changes when a voice agent answers.
Discord TTS bot: add one or build your own
Two ways to get text-to-speech in a Discord voice channel: invite a hosted bot, or build your own with discord.js and Cartesia. Working code included.
Speech-to-speech models vs cascaded voice agents
Speech-to-speech models take audio in and return audio out. How they compare with STT-LLM-TTS voice agents on latency, cost, control and voices.
What is P95 and P99 latency? How to measure it
P95 and P99 latency are the response times 95% and 99% of requests beat. How percentiles work, how to measure them in Python, and why the tail matters.
Twilio pricing 2026: calls, SMS, and AI agents
What Twilio costs in 2026: per-minute voice rates, phone numbers, SMS and carrier fees, SIP trunking, Conversation Relay, and a full AI voice agent bill.
Python speech to text: file, stream, and mic code
Transcribe audio in Python three ways: a recorded file, a real-time stream with turn detection, and a live microphone. Working code, and which library fits.
G.711 codec: μ-law, A-law, and 8 kHz phone audio
G.711 is the codec most phone calls use: 8 kHz, 8 bits per sample, μ-law or A-law. How it works, how to decode it, and what it means for voice agents.
Text to speech with emotion: how to direct it
How emotional text to speech works: emotion read from the text, tags, speaking styles, and prompts across the main APIs, with code and how to test it.
What is DTMF? Tones, frequencies, and voice AI
DTMF is the pair of tones a phone sends when you press a key. How the tones work, how they travel over VoIP, why digits go missing, and what voice agents do.
IVR testing: types, tools, and voice agent tests
How to test an IVR: functional, regression, load, and speech tests for phone menus, plus how to test an AI voice agent whose answers change on every call.
OpenAI Whisper API pricing 2026, model by model
Whisper costs $0.006 a minute. Every OpenAI transcription price in 2026, the February 2027 shutdown, and what to move to before it.
Deepgram pricing 2026: STT, TTS, and agent costs
What Deepgram costs in 2026: per-minute speech-to-text rates, Aura and Flux TTS, the Voice Agent API tiers, add-ons, and the Growth plan.
Conversational AI examples: 12 real deployments
Twelve conversational AI examples running in production: support, healthcare, housing, lending, sales, and games. What each one does and what changed.
Best open-source TTS models, compared by license
Kokoro, Chatterbox, Piper, Orpheus, VibeVoice, XTTS and more: which open-source TTS models you can ship commercially, and how they score.
Conversational AI in banking: customer calls
Conversational AI in banking answers customer calls. Which calls it handles, where the line goes, how to keep card data out of the model, and how to pilot.
Conversational AI in healthcare: patient calls
Conversational AI in healthcare answers patient calls. What it handles well, where the clinical line goes, what HIPAA requires, and how to pilot it.
What is a warm transfer? Warm vs cold transfer
A warm transfer introduces the caller before handing them over; a cold transfer doesn't. How each works, when to use it, and how voice agents hand off.
WebRTC vs WebSocket for real-time voice AI
WebSocket sends ordered messages over TCP. WebRTC sends media over UDP with jitter buffers and loss recovery. Which to use for voice AI, and where.
Contact center automation: what to automate first
Contact center automation, layer by layer: which calls to automate first, what a voice agent needs to resolve them, what it costs, and what to measure.
What is SIP trunking? How it works for voice AI
SIP trunking carries calls between your phone system and a carrier over the internet. How it works, SIP vs VoIP, and connecting a trunk to a voice agent.
What is a mean opinion score (MOS)?
MOS is a 1–5 average of listener ratings. Here's how it's calculated, what a good score is for calls and for TTS, and why two MOS numbers rarely compare.
Conversational AI design for voice agents
Conversational AI design for voice: how to write for the ear, handle turns and interruptions, confirm what matters, and plan the failure paths and handoffs.
What is neural TTS? How neural voices work
Neural TTS uses a trained neural network to turn text into speech. How it works, how it differs from standard voices, and what "neural" costs on a bill.
AI cold calling: what it is and how to build one
AI cold calling is a voice agent that dials, qualifies, and books meetings on its own. What the calls cost, the consent law, and two ways to build one.
On-premise voice AI: what it is and how it works
On-premise voice AI runs speech recognition, language models, and text to speech inside your own network. Compare cloud, VPC, on-prem, and air-gapped options.
AI voice agent: what it is and how to build one
What an AI voice agent is, the pipeline that runs one, and the latency, turn-taking, and testing decisions that decide whether callers stay on the line.
Voice activity detection: VAD for voice agents
Learn how voice activity detection works, how it differs from turn detection, and how to handle pauses and interruptions in a real-time voice agent.
Word error rate: what WER misses in voice AI
Learn what word error rate measures, why it can mislead TTS evaluations, and how to test the speech quality that matters in a live conversation.
What is conversational IVR? Replacing phone menus
Conversational IVR lets callers say what they need instead of pressing keys. How it routes calls, where voice agents fit, and what to test first.
What is an AI call center? How it works and costs
An AI call center uses voice agents to answer and resolve calls. How a call flows, what it costs per minute, and a pilot plan with a test scorecard.
AI receptionist: what to buy, what to build
Compare ready-made AI receptionists with a custom voice agent. Plan scheduling, human handoffs, costs, and a test for your first inbound calls.
SpeechSynthesis API: JavaScript text to speech
How window.speechSynthesis works: methods, utterance rate, pitch and volume ranges, events, and a working JavaScript text-to-speech demo you can copy.
Python text to speech: save audio as a WAV file
Turn text into a WAV file with Python and the Cartesia SDK. Set up an API key, choose a voice, and fix common audio and authentication errors.
Cartesia Is Hiring a SWE to Raise an Army of Claudes
Writing code feels solved. But engineering is not. We're hiring someone to close the gap.
[Engineering]
Get started today
Talk to an expert.
Connect with a member of our team and learn how Cartesia can help you build world-class voice experiences.
Start building.
Access our models via API and bring a voice agent into production in minutes.