BroadRouter

Speech-to-text

Updated

Speech-to-text, also called automatic speech recognition (ASR), is technology that converts spoken audio into written text. A speech-to-text API accepts an audio file or stream and returns a transcript, usually priced per minute of audio. Accuracy is measured with word error rate and varies with accent, dialect and recording quality.

Also called: STT, ASR, automatic speech recognition, transcription API, audio transcription, voice to text

Voice notes, calls and meetings hold information that software cannot search until it is text. Speech-to-text closes that gap.

How it works

Hugging Face defines automatic speech recognition as “the task of transcribing a given audio to text.” A modern ASR model converts the audio into features, then generates the most likely sequence of words.

Through an OpenAI-compatible API, you upload the file as multipart form data to /v1/audio/transcriptions:

curl https://inference.broadnet.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $BROADNET_API_KEY" \
  -F model=broadnet/cohere-transcribe-arabic \
  -F language=ar \
  -F file=@voice-note.wav

The response is JSON with the transcript in text. Providers set limits on file size and formats; OpenAI’s own endpoint accepts files up to 25 MB (its speech to text guide, checked 2026-11-15).

Why it matters

  • Accuracy decides value. A transcript with wrong names or numbers can be worse than none. Accuracy is measured with word error rate, and it drops on accents, dialects, crosstalk and phone audio.
  • Price per hour of audio. STT is billed per audio minute, so volume is easy to forecast: hours of audio times the hourly rate.
  • Privacy. Voice carries names, phone numbers and health or banking details. Ask whether audio and transcripts are kept; see zero data retention.
  • Pipelines. A transcript often feeds an LLM next, for summaries, routing or extraction.

On BroadNet Inference

Model Languages Price per minute Per hour
BroadNet Transcribe Arabic Arabic, including dialects $0.0042 $0.25
BroadNet Transcribe 14 languages $0.002 $0.12

Both accept WAV or MP3 up to 26,214,400 bytes (25 MiB) per file; a larger file returns 413. Send language=ar for Arabic, or it can come back transliterated into Latin letters. BroadNet reports the Arabic model at #2 of 37 (its September 2026 figure; the live board listed it #4 of 39 on 15 November 2026) on the Open Universal Arabic ASR Leaderboard. Audio and transcripts are not retained.

Other providers can cost less. On 2026-11-14, Soniox’s pricing page listed about USD 0.10 per hour for file (async) transcription and USD 0.12 per hour for real-time, and its language list includes Arabic. That is below BroadNet Transcribe Arabic’s $0.25 an hour, so test both on your own audio. Our open speech-to-text API price index lists more than 20 products, with a source and date for each price.

Both models work on the free $5 starting credit, which covers about 1,190 minutes of Arabic audio. Start with the Arabic speech-to-text guide, convert WhatsApp .ogg files with the voice note guide, and compare providers with the buyer’s guide.

Questions

How is speech-to-text accuracy measured?

With word error rate (WER), the share of words a transcript gets wrong against a correct reference. Lower is better. Measure it on your own recordings, because accuracy varies a lot with accent, dialect and audio quality.

How do I transcribe files larger than the size limit?

Split the audio into shorter pieces, ideally at pauses, and transcribe each. Compressing to MP3 also reduces size.

Why does Arabic sometimes come back in Latin letters?

Without a language hint, some models transliterate. On BroadNet Inference, send language=ar for Arabic.

Read as MarkdownAsk ChatGPTAsk ClaudeAsk Perplexity