Text-to-speech API comparison

Choose Amazon Polly for AWS Standard voices at $4 per 1,000,000 characters, ElevenLabs for a vendor-reported ~75 ms Flash v2.5 inference claim, or Deepgram for its documented 745 ms Aura-2 REST example.

Updated · 10 sources · Methodology

NeedPickWhy
AWS Standard voicesAmazon PollyAmazon Polly lists Standard voices at $4 per 1,000,000 billable characters.12
Flash model inferenceElevenLabs~75 ms Flash v2.5 inference, excluding application/network latency, vendor-reported 2026.5
Aura-2 REST executionDeepgram745 ms total in one documented Aura-2 REST example, vendor-reported 2026.6

Performance

ProviderAPI and model IDPublished latency and execution
Amazon Polly1Standard voicesStandard-voice speech synthesis; price is per character1
Google Cloud TTS2Standard voicesStandard-voice synthesis API; separate from Chirp models2
Deepgram3Aura-2745 ms total in one Aura-2 documentation example, vendor-reported 20266
ElevenLabs4Flash v2.5Flash v2.5 ~75 ms inference, excluding application/network latency, vendor-reported 20265

Documented language coverage

Provider / modelModel-specific coverage
Amazon Polly · Standard29 languages and language variants; English and Spanish voices are listed.7
Google Cloud TTS · Standard59 language/region codes in the Standard voice catalogue, including en-US, es-ES and ja-JP.8
Deepgram · Aura-27 languages: en, es, de, fr, nl, it, ja.9
ElevenLabs · Flash v2.532 languages; includes English and Spanish.10

Prices and request cost

ProviderEnglish / 1,000,000 characters1,000,000-character costSpanish / 1,000,000 characters
Amazon Polly1$4$4$4
Google Cloud TTS2$4$4$4
Deepgram3$30$30$30
ElevenLabs4$40$40$40
Cost example: 1,000,000 billable English characters
Amazon Polly$4
Google Cloud TTS$4
Deepgram$30
ElevenLabs$40

Options, strengths and limitations

  1. Amazon Polly

    Amazon Polly provides Standard speech voices through an AWS account.

    English / 1,000,000 characters
    $4
    Example quantity
    1
    Example total
    $4

    Strengths

    • Amazon Polly lists Standard voices at $4 per 1,000,000 characters before free allowances.1

    Limitations

    • The Polly Standard tariff excludes Neural, Long-Form and Generative voice categories.1

    Sources:1

  2. Google Cloud TTS

    Google Cloud TTS provides Standard voices through a Google Cloud account.

    English / 1,000,000 characters
    $4
    Example quantity
    1
    Example total
    $4

    Strengths

    • Google Cloud TTS lists Standard voices at $4 per 1,000,000 characters before free allowances.2

    Limitations

    • Google’s quoted Standard price does not include the separate Chirp voice models.2

    Sources:2

  3. Deepgram

    Deepgram lists Aura-2 at $30 per 1,000,000 characters before free allowances.

    English / 1,000,000 characters
    $30
    Example quantity
    1
    Example total
    $30

    Strengths

    • Deepgram lists Aura-2 at $30 per 1,000,000 characters before free allowances.3

    Limitations

    • Aura-2 costs $30 per 1,000,000 characters, versus $4 for the Standard-voice rows; voice cloning is not part of this quoted rate.3

    Sources:3

  4. ElevenLabs

    ElevenLabs lists Flash v2.5 at $40 per 1,000,000 characters before free allowances.

    English / 1,000,000 characters
    $40
    Example quantity
    1
    Example total
    $40

    Strengths

    • ElevenLabs lists Flash v2.5 at $40 per 1,000,000 characters before free allowances.4

    Limitations

    • Flash v2.5 has a 40,000-character request limit. The $40 per 1,000,000-character rate excludes other voice models and subscription allocations.4

    Sources:4

Methodology and sources

Amazon Polly / Google Cloud TTS / Deepgram / ElevenLabs. The example is 1,000,000 billable characters before free tiers and subscriptions. Prices are normalized per character, not per audio minute; speaking rate and language change audio duration. Standard voices, Aura-2 and Flash/Turbo are different model classes and are not treated as equal-quality outputs. No matched voice-quality, startup or streaming-latency benchmark is available in these pricing sources. Cards use character-price order; the two Standard-voice options tie at $4. Flash v2.5’s inference-only claim and Aura-2’s network-inclusive example measure different intervals and are not a latency ranking. The main budget uses English voices. The Spanish column uses supported Spanish voices in the same selected model classes at the same published character tariffs, not a promise that a single voice switches languages. Published coverage definitions differ: Polly counts languages and variants; Google’s 59 counts distinct language/region codes in Standard rows, counting regional variants separately and each code once. There is no coverage-count ranking.

  1. Amazon Polly · Standard voices specifications and pricing
  2. Google Cloud TTS · Standard voices specifications and pricing
  3. Deepgram · Aura-2 specifications and pricing
  4. ElevenLabs · Flash v2.5 specifications and pricing
  5. ElevenLabs · Flash v2.5 model latency
  6. Deepgram · Aura-2 latency example
  7. Amazon Polly · Standard language and voice catalogue
  8. Google Cloud · Standard voice locale catalogue
  9. Deepgram · Aura-2 voices and languages
  10. ElevenLabs · Flash v2.5 model and language coverage

Draft narration text with Novita LLMs

Use Novita Model APIs to prepare scripts, then send text to your chosen speech provider. The synthesis rates above are third-party tariffs.

Start BuildingExplore Models