Compare Whisper, Deepgram, ElevenLabs, OpenAI TTS, and Google pricing. Enter minutes or characters → see exact monthly cost. Free, no signup.
Select provider · Enter volume · Results update live
| Provider | Type | Price | Quality | 10K min / 1M chars |
|---|---|---|---|---|
| Deepgram Nova-2CHEAPEST STT | STT | $0.0043/min | Excellent | $43 |
| OpenAI Whisper | STT | $0.006/min | Excellent | $60 |
| Google STT | STT | $0.024/min | Good | $240 |
| OpenAI TTSCHEAPEST TTS | TTS | $0.015/1K chars | Good | $15 |
| Google TTS | TTS | $0.016/1K chars | Good | $16 |
| ElevenLabs | TTS | $0.330/1K chars | Best | $330 |
OpenAI Whisper is open source. Self-hosting on a $6/mo Hetzner CX22 handles ~5K minutes/month at near-zero cost. For over 50K min/month, self-hosting saves $300+/mo.
Calculate your STT or TTS API cost for any production workload.
A voice AI customer support assistant handling 5,000 calls/month (avg 4 minutes): STT at OpenAI Whisper = 20,000 mins × $0.006 = $120/month. TTS for responses (avg 200 words = 1,200 chars per call): 5,000 × 1,200 = 6,000,000 chars × $0.015/1K = $90/month. LLM (GPT-5.4 nano) for response generation ≈ $30/month. Total voice pipeline: ~$240/month for 5,000 calls.
How we source STT and TTS API pricing.
Sources: OpenAI Whisper pricing from OpenAI's API pricing page. Deepgram Nova-2 pricing from Deepgram's pricing documentation. Google Cloud STT pricing from Google Cloud's pricing page. OpenAI TTS pricing from OpenAI's API pricing page. ElevenLabs pricing from their plans page. All are pay-as-you-go rates with no volume commitments.
STT pricing model: All STT APIs are billed per minute of audio, rounded up to the nearest second or minute depending on the provider. Whisper and Deepgram round to the nearest second. Google rounds to the nearest 15 seconds. This affects short audio clips significantly — a 10-second clip costs the same as a 15-second clip on Google. The calculator uses per-minute billing for all providers as a conservative estimate.
TTS pricing model: All TTS APIs are billed per 1,000 characters of input text. Characters include spaces and punctuation. A 500-word speech script is approximately 2,800 characters. ElevenLabs charges for the raw character count; OpenAI charges for the tokenized character count (slightly different for non-ASCII text).
Update cadence: STT/TTS pricing is reviewed monthly. ElevenLabs in particular adjusts pricing with plan structure changes; we monitor their pricing page and update within 48 hours of changes.
ElevenLabs at $0.330/1K chars is 22× more expensive than OpenAI TTS Standard ($0.015/1K chars). The quality and naturalness of ElevenLabs voices is significantly better — worth it for customer-facing voice products, audiobooks, or brand voice applications where naturalness is a core feature. For internal tools, notifications, or any automated voice that doesn't need to sound human, OpenAI TTS is more than sufficient.
OpenAI TTS Standard at $0.015 per 1,000 characters is the cheapest quality managed TTS API. Google TTS is similar at $0.016/1K chars. For very large volumes (tens of millions of characters per month), Azure Neural TTS or Amazon Polly can be cheaper with committed-use pricing. Self-hosting Coqui TTS or Bark on a VPS eliminates per-character API costs entirely for teams with DevOps capacity.
Deepgram Nova-2 at $0.0043/minute is the cheapest production-quality STT API. OpenAI Whisper at $0.006/minute is slightly more expensive but offers broader language support and is simpler to integrate (standard REST API). Google STT at $0.024/minute is the most expensive in this comparison but excels at medical/legal vocabulary and multiple speaker diarization. Self-hosted Whisper (open source) is free at the cost of compute.
Approximately 800–1,000 characters per minute of spoken text at normal speaking pace (~150 words/minute). A 10-minute podcast script is roughly 9,000 characters = $0.135 with OpenAI TTS Standard. A 1-minute automated phone greeting (≈ 800 chars) = $0.012. A full audiobook chapter of 5,000 words (≈ 28,000 chars) = $0.42 with OpenAI TTS.
Self-hosting becomes cost-effective at roughly 50,000+ minutes/month for STT. At that volume, OpenAI Whisper API costs $300/month. Self-hosting Whisper on a Hetzner VPS (2 vCPU/4GB) handles ~5,000 minutes/month for $4.50/month in compute. For 50,000 min/month you'd need ~10 servers ($45/month) — a 6.5× cost reduction. Whisper is open source, well-documented, and supports 57 languages. The setup and maintenance overhead is 4–8 hours initially.
Speaker diarization is the process of identifying who said what in a multi-speaker recording (labeling segments as "Speaker 1", "Speaker 2", etc.). Deepgram includes basic diarization in their standard pricing. Google STT charges an additional premium for diarization. OpenAI Whisper API does not natively support diarization — you'd need to use a separate service like pyannote.audio. For meeting transcription with speaker labels, Deepgram is typically the best value.