Free tools Windows power users keep installed
One-click scans. No signup required.
Sarvam AI’s voice-agent strategy is a distribution bet: in a multilingual country where typing Indian scripts can be cumbersome and English is not a universal interface, speaking may be the simplest way to reach customers. The company’s business case is strongest when a voice agent handles a bounded workflow—support, collections, payments or scheduling—through channels people already use.
That does not make voice automatically better than text. Accuracy by language and accent, privacy, fraud controls, telephony reliability, human escalation and the total cost of a completed interaction will determine whether the strategy works.
What Sarvam launched on August 13, 2024
Sarvam introduced voice agents as one part of a broader Indian-language AI stack, not as a standalone chatbot. Its launch suite included the following products:
| Product | Role at launch |
|---|---|
| Sarvam Agents | Multilingual, voice-enabled business agents for customer support, payments, sales and other task-oriented workflows. The launch described telephone, WhatsApp and in-app deployment, with support for 10 Indian languages. |
| Sarvam 2B | An open-source small language model aimed at Indic-language tasks. |
| Shuka 1.0 | An open-source audio-language model. |
| Sarvam Models | Speech recognition, text-to-speech, parsing and related APIs. |
| A1 | A generative-AI workbench for legal users. |
Sarvam announced a starting price of ₹1 per minute for the original agents in 2024. That was a launch-price signal, not a current universal price for every Sarvam voice deployment. Today’s public API sheet prices individual services such as transcription and speech synthesis, while enterprise deployments can have separate terms. The launch announcement is documented at Sarvam’s product post, and the contemporaneous account is in TechCrunch.
Recommended Free Tools
#1 Best Overall
- Your favorite music and content – Play music, audiobooks, and podcasts from Amazon Music, Apple Music, Spotify and others or via Bluetooth throughout your home.
- Alexa is happy to help – Ask Alexa for weather updates and to set hands-free timers, get answers to your questions and even hear jokes. Need a few extra minutes in the morning? Just tap your Echo Dot to snooze your alarm.
- Keep your home comfortable – Control compatible smart home devices with your voice and routines triggered by built-in motion or indoor temperature sensors. Create routines to automatically turn on lights when you walk into a room, or start a fan if the inside temperature goes above your comfort zone.
- Do more with device pairing – Fill your home with music using compatible Echo devices in different rooms, or create a home theatre system with Fire TV.
- Say goodbye to drop-offs and buffering - With eero Built-in, Echo Dot doubles as a mesh wifi extender, adding up to 1,000 sq. ft. of wifi coverage to your existing eero network.
Why voice can be a useful Indian distribution layer
It reduces typing friction
Typing accurately in Devanagari, Bengali, Tamil or another Indian script is not equally easy for every mobile user. Speech lets a customer state a problem without finding a keyboard, switching scripts or spelling a long name and address. Sarvam co-founder Vivek Raghavan has framed speaking in a familiar language as a more natural interface; that is a company thesis, not proof that every Indian user prefers voice.
It reaches beyond English-first interfaces
Indian users routinely move among regional languages, English and code-mixed speech. Sarvam’s documentation specifically discusses building for code-mixed input. A system can identify the language, transcribe it, translate or transliterate where needed, reason over the request and answer in speech. Current documentation lists Saaras v3 speech recognition for 23 languages—22 Indian languages plus English—and Bulbul v3 with more than 30 voices across 11 languages. Those are model-specific coverage claims; support does not imply equal accuracy for every dialect, accent or use case.
Telephone and WhatsApp are already familiar channels
A phone call can work without a new app or a high-end smartphone. WhatsApp can add conversational support inside a channel businesses and customers already understand. But channel availability is not distribution by itself. The business still needs a reason for people to call, reliable carrier infrastructure, authentication, customer education and protection against spam and fraud.
Conversation can complete a task
A text chatbot may answer a frequently asked question. An agent can conduct a dialogue, retrieve a customer record and call a tool to schedule an appointment, create a ticket or initiate a payment workflow. The value comes from completing the job, not from sounding human.
Voice also has clear disadvantages. It is harder to use privately in a crowded place, less searchable than text, awkward for long documents and inaccessible to some people with hearing or speech impairments. Forms, maps, tables and exact account numbers may be better handled visually.
Rank #2
- Your favorite music and content – Play music, audiobooks, and podcasts from Amazon Music, Apple Music, Spotify and others or via Bluetooth throughout your home.
- Alexa is happy to help – Ask Alexa for weather updates and to set hands-free timers, get answers to your questions and even hear jokes. Need a few extra minutes in the morning? Just tap your Echo Dot to snooze your alarm.
- Keep your home comfortable – Control compatible smart home devices with your voice and routines triggered by built-in motion or indoor temperature sensors. Create routines to automatically turn on lights when you walk into a room, or start a fan if the inside temperature goes above your comfort zone.
- Do more with device pairing – Fill your home with music using compatible Echo devices in different rooms, or create a home theatre system with Fire TV.
- Say goodbye to drop-offs and buffering - With eero Built-in, Echo Dot doubles as a mesh wifi extender, adding up to 1,000 sq. ft. of wifi coverage to your existing eero network.
Why enterprise support is the first commercial wedge
Enterprises already spend money on call centres and support operations. That gives a voice agent a defined buyer, a measurable baseline and a narrow workflow in which errors can be controlled. A pilot can track:
- containment and task-completion rates;
- resolution and transfer rates;
- average handling time and abandonment;
- repeat contacts and customer satisfaction; and
- cost per resolved interaction.
Companies also have the material an agent needs: product catalogues, policy documents, scripts, FAQs and CRM records. Connecting that structured information is usually more practical than asking a general consumer chatbot to answer anything.
Where the model can fit
- Financial-services reminders, collections and account support.
- Insurance policy servicing and claim-status requests.
- Telecom troubleshooting and plan changes.
- Consumer-product helplines.
- Healthcare scheduling and information triage, with strict limits on medical advice.
- Government-service access in regional languages.
- Education, agricultural outreach and farmer feedback.
- Religious or subscription-content services.
Sarvam’s current website describes multilingual consumer-loan engagement, insurance-policy calls and farmer-feedback calls. These are company-reported examples, not independently audited performance claims.
The Sri Mandir example
TechCrunch reported Sarvam’s statement that Sri Mandir used an agent in a payment workflow and that more than 270,000 transactions were processed through it. This illustrates the difference between answering questions and participating in a transaction. It does not establish the agent’s share of Sri Mandir’s total transactions, conversion rate, payment-failure rate, customer satisfaction or incremental revenue. “Processed” should not be rewritten as “generated 270,000 successful sales.”
Why Sarvam combines small, specialised and larger models
Routine voice interactions do not always need the largest available model. A smaller or specialised model can reduce inference cost and latency, run in more constrained environments, behave more predictably in a bounded workflow and be tuned for a particular language task. Sarvam said at launch that Sarvam 2B was trained on 4 trillion tokens and used synthetic data because high-quality Indian-language material on the open web is limited. Synthetic data can expand scarce training material, but mistakes and biases in generated text can also be reproduced or amplified. Sarvam’s launch claims that the model was cheaper than comparable systems or better than Meta’s Llama for Indian-language tasks should be treated as company claims unless a benchmark and test conditions are supplied.
Rank #3
- Your favorite music and content – Play music, audiobooks, and podcasts from Amazon Music, Apple Music, Spotify and others or via Bluetooth throughout your home.
- Alexa is happy to help – Ask Alexa for weather updates and to set hands-free timers, get answers to your questions and even hear jokes. Need a few extra minutes in the morning? Just tap your Echo Dot to snooze your alarm.
- Keep your home comfortable – Control compatible smart home devices with your voice and routines triggered by built-in motion or indoor temperature sensors. Create routines to automatically turn on lights when you walk into a room, or start a fan if the inside temperature goes above your comfort zone.
- Do more with device pairing – Fill your home with music using compatible Echo devices in different rooms, or create a home theatre system with Fire TV.
- Say goodbye to drop-offs and buffering - With eero Built-in, Echo Dot doubles as a mesh wifi extender, adding up to 1,000 sq. ft. of wifi coverage to your existing eero network.
The current platform is broader than the original “small model plus voice” description. Sarvam’s catalogue now lists Sarvam-30B and Sarvam-105B chat models alongside speech and language services. The strategic picture is therefore full-stack and model-flexible: use the model and deployment mode that fit the language, latency, risk and cost of a workflow.
What a production voice agent actually contains
A “voice bot” is a chain of systems, each of which can fail:
- Telephone, WhatsApp or in-app audio capture.
- Voice-activity detection, turn-taking and interruption handling.
- Speech-to-text transcription.
- Language identification, with translation or transliteration when required.
- Language-model reasoning and retrieval from approved company content.
- Tool calls to a CRM, payment system, scheduler or ticketing platform.
- Text-to-speech generation and streamed audio playback.
- Logging, monitoring, consent management and human handoff.
Sarvam provides APIs for speech recognition, translation, transliteration, text-to-speech and chat, and publishes guidance on streaming and barge-in at its API overview and Building for India. A fluent response is not evidence that the system heard the request correctly, retrieved the right policy or executed an action successfully.
What current prices do—and do not—tell a buyer
Sarvam’s developer documentation lists component prices rather than a single all-in voice-agent rate:
| Service | Public price shown in the documentation | Qualification |
|---|---|---|
| Speech-to-text | ₹30 per audio hour | Public API rate; enterprise terms may differ. |
| Speech-to-text with diarization | ₹45 per audio hour | Public API rate. |
| Speech-to-text plus translation | ₹30 per audio hour | Public API rate. |
| Bulbul v2 text-to-speech | ₹15 per 10,000 characters | Public API rate. |
| Bulbul v3 text-to-speech | ₹30 per 10,000 characters | Listed as beta pricing. |
| Translation or transliteration | ₹20 per 10,000 characters | Public API rates. |
| Language identification | ₹3.50 per 10,000 characters | Public API rate. |
| Document digitisation | ₹0.50 per page | Maximum 10 pages per job, according to the pricing page. |
See the developer pricing documentation and marketing pricing page. Those public pages also show inconsistent introductory-credit information: the documentation says ₹100 for a new user, while the marketing page says plans start with ₹1,000 in free credits and separately displays a ₹300 starter-credit figure. Buyers should verify the live signup terms rather than budget from either number.
Rank #4
- Meet Echo Dot Max: Experience rich room-filling sound that automatically adapts to your space and fine-tunes playback. Features a built-in smart home hub and Omnisense technology for highly personalized experiences.
- Music to your ears: With nearly 3x the bass versus Echo Dot (2022 release), it fits beautifully in any space, delivering your personal sound stage with deep bass and enhanced clarity. Listen to streaming services, such as Amazon Music, Apple Music, Spotify, and SiriusXM. Encore!
- Do more with device pairing: Connect compatible Echo smart speakers and smart displays in different rooms, or pair with a second Echo Dot Max to enjoy even richer sound. Pair your Echo Dot Max with compatible Fire TV devices to create a home theater system that brings scenes to life.
- Simple smart home control: Set routines, pair and control lights, locks, and thousands of smart home devices that work with Alexa without needing a separate smart home hub. With Omnisense technology, you can activate routines via temperature or presence detection.
- Say goodbye to drop-offs and buffering - With eero Built-in, Echo Dot Max doubles as a mesh wifi extender, adding up to 1,000 sq. ft. of wifi coverage to your existing eero network.
The real cost of a completed interaction is closer to:
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →telephony + speech recognition + translation or language processing + language-model tokens + text-to-speech + retrieval and databases + monitoring + human escalation + integration.
The original ₹1-per-minute announcement cannot be compared directly with today’s component API rates without knowing what was included.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Sovereign AI gives Sarvam a strategic context
Sarvam’s pitch aligns with India’s effort to build domestic language, speech and foundation-model capabilities. The IndiaAI Mission and related government documents identify multilingual, speech and voice-model work, including Sarvam’s participation, as part of that effort. Relevant goals include local-language public services, domestic infrastructure, data-residency options and models tuned to Indian legal and administrative contexts. Government alignment can provide legitimacy, procurement opportunities and distribution, but it does not guarantee product-market fit. Public-sector buying cycles are long, requirements are complex and deployments can be politically sensitive.
“Sovereign” also does not automatically mean that data never leaves India. A buyer must inspect the specific hosting location, subprocessors, retention, deletion, access controls and private-deployment terms.
Best Value
- Alexa can show you more - Echo Show 5 includes a 5.5” display so you can see news and weather at a glance, make video calls, view compatible cameras, stream music and shows, and more.
- Small size, bigger sound – Stream your favorite music, shows, podcasts, and more from providers like Amazon Music, Spotify, and Prime Video—now with deeper bass and clearer vocals. Includes a 5.5" display so you can view shows, song titles, and more at a glance.
- Keep your home comfortable – Control compatible smart devices like lights and thermostats, even while you're away.
- See more with the built-in camera – Check in on your family, pets, and more using the built-in camera. Drop in on your home when you're out or view the front door from your Echo Show 5 with compatible video doorbells.
- See your photos on display – When not in use, set the background to a rotating slideshow of your favorite photos. Invite family and friends to share photos to your Echo Show. Prime members also get unlimited cloud photo storage.
Where voice agents fail
Recognition and language quality
- Regional accents, dialects and noisy streets can lower transcription accuracy.
- Users may switch languages mid-sentence.
- Names, addresses, amounts and product codes are especially error-prone.
- Children, older users and people with speech impairments may be harder to recognise.
- Text-to-speech may mispronounce local names and place names.
Ask for word-error rates and task results for each target language and accent, not one aggregate “language support” number.
Latency and conversation control
Translation, retrieval, safety checks and tool calls improve control but add delay. Production systems need streaming, interruption (“barge-in”) handling, sensible timeout rules and a text or human fallback. Otherwise the agent can talk over the user, terminate a turn too early or enter a repetition loop.
Transactions, fraud and high-risk advice
Never let a language model alone authorise a payment. Confirm the recipient, amount and purpose in a structured step; authenticate outside the model; record the user’s confirmation and the tool response; and design for retries and partial failures. Repeated recognition failures, identity uncertainty, complaints, legal threats, medical or financial advice and emotional distress should trigger explicit escalation rules.
Privacy and compliance
Recordings may contain financial and other sensitive information. Organisations need retention and deletion policies, role-based access, disclosure that the caller is speaking with AI and sector-specific consent procedures. Cross-border processing and vendor access require contract and security review.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11How to evaluate a vendor before committing
- Use representative recordings. Test real accents, background noise, code-switching and the exact languages your customers use.
- Measure each language separately. Record transcription errors, task completion, latency, abandonment and repeat contacts.
- Test interruption and failure recovery. Check barge-in, long pauses, dropped calls, duplicate requests and a text or human fallback.
- Constrain tool use. Require authentication, confirmation and idempotent payment or CRM operations outside the model.
- Price the completed task. Include carrier fees, model calls, infrastructure, monitoring, integration and human review.
- Set data terms in writing. Confirm hosting, retention, deletion, training use, incident response and access controls.
- Pilot with a staffed escalation path. Start with a bounded workflow and audit high-risk conversations before expanding.
The bottom line
Sarvam’s bet makes sense when voice is treated as a distribution and workflow layer for Indian-language AI—not as a novelty interface. India’s language diversity, typing friction and existing phone and WhatsApp habits create a credible opening. The hard part is converting that opening into reliable, safe and economical task completion. Language quality by dialect, low latency, secure tool use, human handoff, enterprise integration and fully loaded cost will decide whether the 2024 thesis scales.
Frequently Asked Questions
Is Sarvam’s ₹1-per-minute price still the standard price for voice agents?
No. ₹1 per minute was the starting price announced for Sarvam Agents in August 2024. Current public documentation prices speech and language APIs separately, and enterprise deployments may have different terms.
Does support for 23 languages mean equal performance in every language?
No. Saaras v3 is documented for 23 languages, but coverage does not establish equal accuracy across dialects, accents, noise conditions or code-mixed speech. Buyers should test their own recordings.
Can a voice agent safely approve payments by itself?
It should not. Payment authorisation, authentication, amount confirmation and transaction logging must be handled by structured systems outside the language model, with escalation for uncertainty or failure.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
