A phone call carries mu-law audio at 8 kHz, and Bland Speech, Deepgram, and Rime all return it, so the audio format is not what separates the three. What does is how each one bills, how each describes its own speed, and how the voices fared in Design Arena's blind listening results of 4 August 2026, where one of the three was scored.
Bland Speech v3 renders natively at 48 kHz and steps down to 8 kHz mu-law for a phone line, which is the same job Deepgram's Flux TTS and Rime's Coda do. Rime bills by the minute, Deepgram and Bland by the character, and neither Deepgram nor Rime was in Design Arena's 4 August 2026 listening test. The text to speech comparison hub lists the other providers.
Which phone call voice is best: Bland Speech, Deepgram, or Rime?
For phone calls where a caller should hear a person, Bland Speech is the pick of the three as of November 2026, because Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, and neither Deepgram nor Rime was in those results. Deepgram is the pick for speech recognition and synthesis from one vendor and for hosting text to speech itself, and Rime is the pick for a regulated enterprise that wants Rime's Forward Deployed Engineer and its own VPC or on-premise engine.
What these rows mean
- Audio Realism Bench: Design Arena's blind listening test, where listeners pick which of two clips of the same line sounds more like a real person, and the votes become an Elo rating, a score that rises each time a voice is picked.
- First audio, as published: The time from sending text to hearing the first sound. P50 is the median, so half of requests start sooner. Each number is the company's own claim, and none is a race Bland ran.
- Price: Rime says about 1,000 characters make a minute of speech, which puts Coda v2 near $50 per million characters and Mist v3 near $30 on Flex.
- Phone audio: Mu-law is 8-bit compressed audio at 8 kHz, used on North American and Japanese phone lines and by Twilio. A-law is the other common phone encoding.
- Compliance: Each cell is the company's own description. Bland's cell describes Bland, the company, and not the Bland Speech API as a separate certification.
What does a phone line do to each voice, and what do the speed claims count?
A phone line takes the voice model's output down to 8 kHz, and each of the three produces that format directly, so the audio reaching a caller is mu-law from all of them. The speed claims are where the three diverge the most in how they are written.
What do the first-audio numbers count?
Deepgram says Flux TTS responds in as low as 80ms at full 24kHz fidelity, even under production load. Bland publishes 199ms to first audio on a streamed request, counted to the first audio chunk rather than the finished file. Rime publishes two pictures. Rime gives Coda v2 a median of 30 ms and a 95th percentile of 60 ms, measured at the server without network time. Rime also gives Coda a median of 96 ms for one request at a time and 150 ms at 12 concurrent requests, and says the cloud adds about 25 to 50 ms of network.
Those are three definitions. Rime's 30 ms is the smallest figure but leaves out network time, Deepgram's 80ms comes next, and Bland's 199ms is the largest. A phone agent also waits on speech recognition and a language model before any voice is generated, so the figure is one part of the delay a caller hears.
Where do Deepgram and Rime fall short next to Bland Speech?
Deepgram falls short on its two-generation lineup, and Rime on how its speed and cloning are sold.
Where Deepgram falls short
Deepgram sells Flux TTS, which it recommends for all new builds, at $0.045 per 1,000 characters on pay as you go and on Growth, and Aura-2, which it calls a previous-generation model, at $0.030 on pay as you go and $0.027 on Growth. Deepgram says SSML, the markup language for speech tweaks, is stripped from input instead of supported, and a delivery dial from calm to animated, -2 to 2, is marked beta.
Pros
- First audio as low as 80ms at 24 kHz under production load, as Deepgram publishes it
- Mu-law and A-law output at 8 or 16 kHz, with a Twilio guide that requests mu-law at 8 kHz to match Twilio Media Streams
- Self-hosting, private cloud, and on-premise deployment under Deepgram's terms
- A $200 free credit with no card and no expiry
Cons
- Not in the 4 August 2026 Audio Realism Bench results
- Flux TTS costs $45 per million characters, above Bland Speech's $40
- Aura-2, the cheaper voice at $30, is previous generation
- SSML is stripped, and the delivery dial is beta
Use Deepgram when:
- A team wants speech recognition and speech synthesis from one vendor
- A security team will host text to speech itself, in a private cloud or on-premise
Where Rime falls short
Rime sells conversational models for production voice agents, IVR, and telephony in regulated industries. Coda, released in May 2026, is its flagship text to speech model, and Mist v3, released in March 2026, is its low-latency one. Rime retired Arcana on 15 August 2026 and retires Mist v1 and v2 on 3 December 2026, so a team on an older model is on a clock.
Voice cloning is an enterprise order: Rime says an enterprise clone is trained on Coda or Mist, and 30 to 60 minutes of clean recordings make a basic one. Rime measures its 30 ms at the server without network, and its latency benchmark shows Coda rising from 96 ms to 150 ms at 12 concurrent requests.
Pros
- A first-audio median of 30 ms for Coda v2, measured at the server
- HTTP, WebSocket, and gRPC streaming, with word-level timestamps on WebSocket and gRPC
- Dedicated VPC or fully on-premise deployment for enterprise, with a Forward Deployed Engineer and Linguist in the SLA
- SOC 2 Type 2 and HIPAA compliance, with a BAA available
- $50 of free usage every month
Cons
- Not in the 4 August 2026 Audio Realism Bench results
- Two different first-audio figures for Coda, 30 ms and 96 ms at the median
- Cloning is an enterprise order from 30 to 60 minutes of recordings
- Each Coda voice serves one language, across 10 cloud languages
- Mist v1 and v2 retire on 3 December 2026
Use Rime when:
- A regulated team wants a vendor engineer and a linguist in the SLA, and an engine inside its own VPC or on-premise
- A team will meter text to speech by the minute, at $0.050 for Coda v2 or $0.030 for Mist v3
What do Bland Speech, Deepgram, and Rime cost?
Bland Speech costs $40 per million characters, Deepgram $45 on Flux TTS or $30 on Aura-2, and Rime $50 per million characters on Coda v2 or $30 on Mist v3 at Flex rates, by Rime's own figure of about 1,000 characters a minute.
Bland Speech pricing
Bland Speech costs $0.04 per 1,000 characters, which is $40 per million, pay as you go and the same in the studio and the API, with no monthly plan. A new account gets 50,000 characters free once, and credits don't expire. Enterprise pricing is custom, set with Bland's sales team.
Deepgram and Rime pricing
Deepgram charges $0.0450 per 1,000 characters for Flux TTS, $0.030 for Aura-2, and $0.0150 for Aura-1 on pay as you go, with $200 of free credit that doesn't expire. Rime has no monthly fee on Flex and bills $0.050 a minute for Coda v2, $0.030 for Mist v3, and $0.080 for Prism, its speech-to-speech model. Rime's Scale plan costs $100 a month with $100 of credit and takes 20% off every rate, so Coda v2 is $0.040 a minute. Rime's Free plan includes $50 of usage a month, which at the Coda v2 rate is 1,000 minutes.
What does a month of phone speech cost on each?
Worked from published prices, at Rime's 1,000 characters a minute:
- 1 million characters, about 1,000 minutes: Bland Speech $40. Deepgram Flux TTS $45, or $30 on Aura-2. Rime Coda v2 $50 on Flex, or $30 on Mist v3.
- 10 million characters, about 10,000 minutes: Bland Speech $400. Deepgram Flux TTS $450, or $300 on Aura-2. Rime Coda v2 $500 on Flex or $400 on Scale, and Mist v3 $300 on Flex.
Deepgram's Aura-2 and Rime's Mist v3 cost less than Bland Speech per character, and Deepgram's Aura-1 at $15 per million costs less again. Bland Speech's rate stays $40 at any volume.
Which sounds more human on a phone call, Bland Speech, Deepgram, or Rime?
Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365, where the human recording scored 1500. Deepgram and Rime weren't in those results. Design Arena describes the result in its own words: "Second only to human speech is Bland Speech v3 by Bland AI, with an Elo of 1365."
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
- Not rankedDeepgram's own voicesNot ranked
Ranks are among 13 AI voices, and Deepgram's own voices weren't ranked. Real human recordings scored 1500, at the right edge of this scale. Bars start at 600, not zero. Source: Design Arena's published results.
The test used one American-English female voice and one male voice per model, chosen as its most naturally conversational voice. Its categories include phone agents, whose prompts tend to be short single turns, and it leaves acting-heavy speech out of scope.
Why is Bland Speech the stronger pick for a phone voice?
Bland Speech is the pick of the three when the voice is what a caller judges the call by. Bland says Bland Speech "was built for phone calls, where polish reads as fake," and Bland runs its own model and speech instead of wrapping a third party's. Performance tags in square brackets, such as [laughs], [sighs], or [say warmly], shape how the words after them are delivered, and an instant clone takes one clean sample of about 10 seconds.
A professional clone takes 30 minutes or more of audio after Bland's consent and rights review, and every clone is private to the customer's organization, with the customer confirming the voice is theirs to use. Bland Speech has a first-party Pipecat service and a LiveKit plugin, the endpoints are covered under the text to speech API, and Bland Speech has the studio.
Bland, the company behind the model, is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted 17 November 2026, and its trust page says speech synthesis runs inside the FedRAMP boundary.
Pros
- Ranked first of 13 AI voices on Audio Realism Bench on 4 August 2026
- One standard rate of $0.04 per 1,000 characters, at any volume
- Mu-law at 8 kHz, and PCM from 8 to 48 kHz
- Performance tags, and an instant clone from about 10 seconds of audio
- A first-party Pipecat service and a LiveKit plugin
Cons
- More than a hobby project needs, because Bland Speech is built for production calls
Bland Speech vs Deepgram and Bland Speech vs Rime cover each pair on its own, and Deepgram Aura vs Rime covers the pair without Bland. The rival comparisons page has every pairing.
Frequently asked questions
Which voice is best for phone calls?
Bland Speech is the strongest pick for phone calls on the evidence available: Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, while Deepgram and Rime weren't in those results. Deepgram and Rime each suit a team with a specific need, such as 80ms published first audio or a vendor engineer in the SLA.
Do all three return mu-law for phone lines?
Yes. Bland Speech returns mu-law at 8 kHz, Deepgram returns mu-law and A-law at 8 or 16 kHz, and Rime returns G.711 mu-law (audio/PCMU) with an 8 kHz sampling rate. Mu-law is the compressed 8-bit format used on North American phone lines and by Twilio.
How is Rime priced compared with Bland Speech?
Rime bills by the minute and Bland Speech by the character. Rime's Coda v2 is $0.050 a minute on Flex, which is near $50 per million characters at Rime's own figure of about 1,000 characters a minute, and Mist v3 is $0.030 a minute. Bland Speech is $40 per million characters.
Which of the three publishes the lowest first-audio figure?
Rime publishes the lowest, 30 ms at the median for Coda v2, measured at the server without network, against Deepgram's as low as 80ms for Flux TTS and Bland's published 199ms. Each figure uses its company's own definition, and Rime also publishes 96 ms at the median for Coda with one request at a time.
Can Bland Speech and Rime clone a voice for a phone agent?
Both can, with different routes. Bland Speech makes an instant clone from one clean sample of about 10 seconds and a professional clone from 30 minutes or more of audio, after Bland's consent and rights review. Rime's cloning is an enterprise order, trained on Coda or Mist from 30 to 60 minutes of clean recordings for a basic clone.
