Tanpura Speech (Demo)

domain verifiedVendor claimDemo

Indian-language voice AI with published code-switching accuracy.

Built for markets where callers move between two languages inside a sentence. Publishes a code-switching figure separately from blended accuracy, which is unusual and is the number worth asking every vendor for.

tanpura-speech.exampleIN

99%

profile complete

Sample data — this company does not exist

Everything on this page is invented so the matching, comparison and review features can be shown working end to end. No figure here describes any real vendor. The genuine profiles in this directory are marked compiled by us and have most of these fields deliberately empty, because nobody publishes them.

Time, effort and pilot terms

The fields buyers ask for most, and vendors publish least

Median sales response: 8 hours
Time to a dialable prototype: 7 working days
Time to production: 6 weeks
Your team's effort to prototype: 14 hours
Who configures it: Guided
Training data required: Documents
Pilot available: Yes
Pilot length: 30 days
Pilot cost: Paid, credited back
Contract required before pilot: No

Languages

Proficiency and code-switching are scored separately

Hindiहिन्दीNativeCode-switchingVerified
Indian EnglishEnglish (India)NativeCode-switchingVerified
MarathiमराठीNativeCode-switchingVerified
Tamilதமிழ்NativeCode-switchingVendor attested
TeluguతెలుగుNativeno code-switchingVendor attested
BengaliবাংলাAccentedno code-switchingVendor claim
Kannadaಕನ್ನಡAccentedno code-switchingVendor claim

Capabilities

CONVERSATION

Outbound diallingVerified
Inbound answeringVerified
Multi-turn dialogueVerified
Barge-in / interruptionVendor attested
Warm human handoffVerified

INTELLIGENCE

Code-switchingVerified
Accent robustnessVerified
Intent detectionVerified
Sentiment detectionVendor claim

COMPLIANCE

DND / DLT complianceVerified
Consent captureVendor attested
PII redactionVendor attested

OPERATIONS

CRM write-backVendor attested
Campaign managementVendor attested

TELEPHONY

Call recordingVerified

Published metrics

Every number here carries its sample size

Code-switch accuracy91.3%n=8,820 · 2026-05Verified
Hindi-English and Marathi-English switches, held-out production sample.
Intent accuracy restricted to utterances containing a mid-sentence language switch, such as Hindi-English. Commonly distorted by: Not measuring it at all, and quoting the blended figure instead. If a vendor sells into India and has no number here, that is itself the finding.
Containment rate64.8%n=31,400 · 2026-05Verified
Collections reminders across four NBFC clients, all conversations reaching the bot.
Share of conversations fully resolved by the bot with no transfer to a human and no callback within 24 hours. Commonly distorted by: Excluding abandoned calls from the denominator, which can move the figure 15–20 points without changing anything real.
Speech recognition WER11.7%n=214,000 · 2026-05Vendor attested
Hindi telephony audio at 8 kHz against human transcription.
Word error rate of the speech-to-text layer against human transcription, on production audio in the stated language. Commonly distorted by: Quoting a benchmark figure on clean studio audio. Telephony audio is 8 kHz and noisy; the real number is usually far worse.
Barge-in handling96.0%n=120 · 2026-05Vendor claim
Barge-in events in a single pilot week.
Share of caller interruptions where the bot stopped speaking within 300 ms and responded to what was said. Commonly distorted by: Counting the stop but not whether the bot then answered the right question.
Sample too small to be meaningful for this metric — shown struck through.