Skip to main content

Koyal TTS by Plivo

Compare Koyal with other text to speech APIs

Hear the same line on Koyal and the leading text to speech APIs, then see what each one costs.

The line

One take each · default settings · recorded 18 Nov 2026

Your verification code is 482916. Again, that's 4 8 2 9 1 6. It expires in 10 minutes.

  • Koyal · Garrett

    First audio
    200 ms
    Per 1K chars
    $0.0065
    0:00 / 0:11
  • Eleven v4 Turbo · Roger

    First audio
    193 ms
    Per 1K chars
    $0.0400
    0:00 / 0:10
  • TTS v2 · Adrian

    First audio
    285 ms
    Per 1K chars
    $0.0120
    0:00 / 0:11
  • Fish Audio

    Drama 3 · Adrian

    First audio
    271 ms
    Per 1K chars
    $0.0500
    0:00 / 0:14
  • Sonic 3.6 · Daniel

    First audio
    286 ms
    Per 1K chars
    $0.0500
    0:00 / 0:09
  • GPT-Realtime-2.1 mini · Cedar

    First audio
    not published
    Per 1K chars
    $0.0242
    0:00 / 0:08
  • 3.8 Flash-Lite TTS · Puck

    First audio
    not published
    Per 1K chars
    $0.0091
    0:00 / 0:09
  • TTS-2 Flash · Dennis

    First audio
    67 ms
    Per 1K chars
    $0.0150
    0:00 / 0:08
  • Flux TTS · Miles

    First audio
    215 ms
    Per 1K chars
    $0.0450
    0:00 / 0:10
  • xAI

    Grok TTS · Altair

    First audio
    316 ms
    Per 1K chars
    $0.0150
    0:00 / 0:09

Language

The line

Your loan of ₹12,45,000 is approved. The first EMI of ₹18,750 is due on 5 December.

  • Koyal · Nisha

    First audio
    200 ms
    Per 1K chars
    $0.0065₹0.65
    0:00 / 0:11
  • Eleven v4 Turbo · Anika

    First audio
    193 ms
    Per 1K chars
    $0.0400₹3.77
    0:00 / 0:10
  • Bulbul V4 Flash · Simran

    First audio
    under 250 ms
    Per 1K chars
    $0.0212₹2.00
    0:00 / 0:10
  • Lightning v3.1 Pro · Zariya

    First audio
    338 ms
    Per 1K chars
    $0.0195₹1.84
    0:00 / 0:08
  • Sonic 3.6 · Kavita

    First audio
    286 ms
    Per 1K chars
    $0.0500₹4.71
    0:00 / 0:09
  • TTS v2 · Priya

    First audio
    285 ms
    Per 1K chars
    $0.0120₹1.13
    0:00 / 0:09
  • Falcon 2.1 · Isha

    First audio
    137 ms
    Per 1K chars
    $0.0100₹0.94
    0:00 / 0:11

One take each · default settings · recorded 17 and 18 Nov 2026

cost

What each one costs

The price of the same amount of speech on each: a minute, an hour, or a month's worth of characters.

Cost comparison

Price for

  • Koyal 65%+ lower

    $0.0065 for the amount selected

  • Eleven v4 Turbo

    $0.0400 for the amount selected

  • TTS v2

    $0.0120 for the amount selected

  • Fish Audio

    Drama 3

    $0.0500 for the amount selected

  • Sonic 3.6

    $0.0500 for the amount selected

  • GPT-Realtime-2.1 mini

    $0.0242 for the amount selected

  • 3.8 Flash-Lite TTS

    $0.0091 for the amount selected

  • TTS-2 Flash

    $0.0150 for the amount selected

  • Flux TTS

    $0.0450 for the amount selected

  • xAI

    Grok TTS

    $0.0150 for the amount selected

Pay-as-you-go prices from each provider's own website, 18 Nov 2026. A minute of speech is counted as 1,000 characters. Koyal is the lowest price here, and more than 65% lower than ElevenLabs, Fish Audio, Cartesia, OpenAI and Deepgram. How we compared

cost

What each one costs

The price of the same amount of speech on each: a minute, an hour, or a month's worth of characters.

Cost comparison

Price for

  • Koyal 65%+ lower65%+ lower

    $0.0065₹0.65 for the amount selected

  • Eleven v4 Turbo

    $0.0400₹3.77 for the amount selected

  • Bulbul V4 Flash

    $0.0212₹2.00 for the amount selected

  • Lightning v3.1 Pro

    $0.0195₹1.84 for the amount selected

  • Sonic 3.6

    $0.0500₹4.71 for the amount selected

  • TTS v2

    $0.0120₹1.13 for the amount selected

  • Falcon 2.1

    $0.0100₹0.94 for the amount selected

Pay-as-you-go prices from each provider's own website, 18 Nov 2026. A minute of speech is counted as 1,000 characters. Koyal is the lowest price here, and more than 65%65% lower than ElevenLabs, Sarvam and Cartesia. How we compared

time to first audio

How soon each one starts speaking

Time to first audio is the wait between sending text and hearing sound. On a call, it is the pause before the agent replies.

Time to first audio

Milliseconds, lower is faster

  • Koyal

    200 ms

  • Eleven v4 Turbo

    193 ms

  • TTS v2

    285 ms

  • Fish Audio

    Drama 3

    271 ms

  • Sonic 3.6

    286 ms

  • GPT-Realtime-2.1 mini

    not published

  • 3.8 Flash-Lite TTS

    not published

  • TTS-2 Flash

    67 ms

  • Flux TTS

    215 ms

  • xAI

    Grok TTS

    316 ms

Coval benchmark, 30-day average. Koyal: Plivo's figure.

How we compared, and sources

The models
Each provider's latest model for real-time voice agents, named in each row.
The price
The pay-as-you-go price on each provider's own website, read on 18 Nov 2026, or its entry plan where it sells plans only. Soniox, OpenAI and Gemini price by tokens or by the hour, so their rates are worked out per character; each row on the right says how. Plan, volume and negotiated rates are not included.
Time to first audio
From Coval's benchmark where it lists the model: a 30-day average, measured from the US to the first audible sample. OpenAI and Gemini publish no figure and are not on the benchmark. Koyal's is Plivo's own. They are not all measured the same way.
The recordings
Made on 18 Nov 2026 with the same text Koyal was sent. One take per line, no retakes, on each provider's default settings, with a voice matched to Koyal's by gender and accent and chosen before listening. OpenAI's model is conversational, so it was given one instruction: to read the text exactly as written. Every take, Koyal's included, is set to the same loudness; nothing else is changed. Recorded so far: ElevenLabs on Eleven v4 Turbo, Soniox on TTS v2, Fish Audio on Drama 3, Cartesia on Sonic 3.6, OpenAI on GPT-Realtime-2.1 mini, Gemini on 3.8 Flash-Lite TTS, Inworld on TTS-2 Flash, Deepgram on Flux TTS and xAI on Grok TTS.
time to first audio

How soon each one starts speaking

Time to first audio is the wait between sending text and hearing sound. On a call, it is the pause before the agent replies.

Time to first audio

Milliseconds, lower is faster

  • Koyal

    200 ms

  • Eleven v4 Turbo

    193 ms

  • Bulbul V4 Flash

    under 250 ms, the provider's own figure

  • Lightning v3.1 Pro

    338 ms

  • Sonic 3.6

    286 ms

  • TTS v2

    285 ms

  • Falcon 2.1

    137 ms

Coval benchmark, 30-day average. * Provider's own published figure. Koyal: Plivo's figure.

How we compared, and sources

The models
Each provider's latest model for real-time voice agents, named in each row.
The price
The pay-as-you-go price on each provider's own website, read on 18 Nov 2026, or its entry plan where it sells plans only. Soniox prices by tokens or by the hour, so its rate is worked out per character; each row on the right says how. Dollar prices are shown in rupees, and Sarvam's rupee price in dollars, at ₹94.26 to the dollar, the mid-market rate on 13 Nov 2026. Koyal has a list price in each currency. Plan, volume and negotiated rates are not included.
Time to first audio
From Coval's benchmark where it lists the model: a 30-day average, measured from the US to the first audible sample. It does not list Sarvam, so that is the provider's own published figure, marked *. Koyal's is Plivo's own. They are not all measured the same way.
The recordings
Made on 17 and 18 Nov 2026 with the same text Koyal was sent. One take per line, no retakes, on each provider's default settings, with a voice in the line's language, matched to Koyal's by gender, and chosen before listening. Where a provider has no such voice, its nearest one reads the line: a male voice reads Sarvam's Tamil line and Smallest.ai's Bengali lines that Koyal reads in a female voice; Murf's Telugu lines, one Tamil line and one Kannada line are read by voices from another Indian language that Murf lists for that language; Cartesia's Bengali voices are from Bangladesh; and Soniox's two Indian voices speak every language. Sarvam's handling of numbers and mixed-language text was switched on, as its docs advise. Every take, Koyal's included, is set to the same loudness; nothing else is changed. Recorded so far: ElevenLabs on Eleven v4 Turbo, Sarvam on Bulbul V4 Flash, Smallest.ai on Lightning v3.1 Pro, Cartesia on Sonic 3.6, Soniox on TTS v2 and Murf on Falcon 2.1.
what to listen for

Where most voices slip up

Most voices sound fine on a simple sentence. These four are where they go wrong on a real call, and each one is in a line you can play above.

  • Codes and digits

    A code is for typing in. It should come out one character at a time, at a pace you can follow. Listen for digits run together into one number, rushed, or dropped.

    Typed 482916

    Listen for four, eight, two, nine, one, six

    Go to the line: Verification code

  • Names

    A name said wrong is the quickest way to sound like a machine. Listen for long names, and names from other languages, said in one go with no pause to work them out.

    Typed Mr. Nguyen, Dr. Kowalczyk

    Listen for each name said once, without spelling it out

    Go to the line: Person names

  • Amounts and dates

    Money and dates have their own way of being read aloud. Listen for the dollars, the cents and the date said the way a person would say them, not read out symbol by symbol.

    Typed $1,249.50

    Listen for one thousand two hundred forty-nine dollars and fifty cents

    Go to the line: Amount and card digits

  • Word for word

    The words you send should be the words that are spoken. Listen for a word that was added, skipped or swapped, and for a sentence that trails off before its end.

    Typed Press 1 to pay now with the card ending in 4821.

    Listen for every word, in order, to the last digit

    Go to the line: Amount and card digits

what to listen for

Where most voices slip up

Most voices sound fine on a simple sentence. These four are where they go wrong on a real call, and each one is in a line you can play above.

  • Codes and digits

    A code is for typing in. It should come out one character at a time, at a pace you can follow. Listen for digits run together into one number, rushed, or dropped.

    Typed KA 05 MJ 4821

    Listen for K A, zero five, M J, four eight two one

    Go to the line: OTP and number plate

  • Names

    A name said wrong is the quickest way to sound like a machine. Listen for long names, and names from other languages, said in one go with no pause to work them out.

    Typed Siddharth Chatterjee, Visakhapatnam

    Listen for each name said once, without spelling it out

    Go to the line: Long names

  • Lakh and crore

    Indian amounts are grouped in lakh and crore, and a caller expects to hear them that way. Listen for the grouping, and for the rupee said as a word and not skipped.

    Typed ₹2,35,000

    Listen for two lakh thirty-five thousand rupees, not two hundred and thirty-five thousand

    Go to the line: Loan and EMI

  • Hindi with English

    People switch between Hindi and English in the middle of a sentence. Listen for the English words keeping their English sound inside a Hindi line, in the same voice, with no pause at the switch.

    Typed loan approve हो गया है

    Listen for loan and approve said as English, the rest as Hindi

    Go to the line: Loan and EMI

faq

Frequently asked questions

Are these the real prices?

They are the pay-as-you-go prices on each provider's own website, read on 18 Nov 2026, or the entry plan where a provider sells plans only. Plan, volume and negotiated rates are not included. Soniox, OpenAI and Gemini price by tokens or by the hour, so their figures are worked out per character.

Why per character and not per minute?

Text to speech APIs bill for the characters you send, and Koyal does too. About 1,000 characters make a minute of speech, so the price for 1,000 characters is roughly the price of a minute.

Are the time to first audio figures measured the same way?

No. Most are from Coval's benchmark, which measures from the US. OpenAI and Gemini publish none. Koyal's is Plivo's own. For a real answer, time each API from where your agents run.

How were the recordings made?

On 18 Nov 2026, each provider was sent the same text as Koyal: one take per line, no retakes, default settings, and a voice matched to Koyal's by gender and accent. Every take is set to the same loudness and nothing else is changed.

Why are some providers not in the recordings?

So far this page has recordings from ElevenLabs, Soniox, Fish Audio, Cartesia, OpenAI, Gemini, Inworld, Deepgram and xAI. To compare any other provider by ear, send it the same lines and listen to the codes, names and amounts.

get started

Put Koyal's voices in your agents

Sign up for an API key and every Koyal voice.